English version below ↓

Índice.

1. El Precio, en Contexto

2. Encontrando la Rentabilidad Esperada a Partir del Cost of Equity

3. ¿Es Eso Mucho o Poco? El Dato Incómodo

4. Entonces, ¿Por Qué Sigo Comprando?

5. Discounted Cash Flow Para Comparar con la Competencia

6. ¿Entonces, Por Qué MCD y no QSR?

7. Conclusión y Cuidadín

Hoy quiero dejar un poco de lado la IA, el petróleo, los chips y el SaaSmageddon, para buscar un poco de paz y tranquilidad 🧘Eso es justo lo que siento cuando entro a un establecimiento de McDonald’s sabiendo que el 4% de mi cartera está en esta acción. Me siento en una mesa, y busco la oferta del día en la aplicación. Y por 3~4 euros, me como una hamburguesa pequeña y unas patatas.

Cuando voy de viaje a otro país, a menudo tengo problemas para comunicarme porque mi inglés es nefasto. Y probablemente me pierda y no sepa utilizar el transporte público porque tengo el instinto de supervivencia de un lemming.

Pero sé que allí me voy a encontrar un McDonald’s donde podré comer y descansar un rato. Un refugio. Un hogar. Y un pellizquito de ese refugio, es mío.

Por eso sigo acumulando la acción que más tranquilidad me da en mi cartera (con permiso de Berkshire). Y aunque tengo bastante ya , estoy deseando que siga cayendo para seguir acumulando. Voy a explicar los números detrás de esa tranquilidad.

El Precio, en Contexto

MCD cotiza ahora mismo con un PER de entorno a 22-23x. Eso es entre un 10% y un 13% por debajo de su media de los últimos 3, 5 y 10 años, que ronda las 25-26x. No es un chollo histórico tipo covid-crash, pero sí es de los múltiplos más bajos que ha tenido la acción en bastante tiempo.

Un PER 22 en una empresa que crece ingresos al 3-4% anual no es barato en términos absolutos. Es barato en términos relativos a su propia historia.

Encontrando la Rentabilidad Esperada a Partir del Cost of Equity

Puesto que McDonald’s es una dividendera clásica, voy a empezar realizando un Dividend Discount Model. En este caso la idea no es utilizar el cost of equity consensuado por los analistas, sino forzar un cost of equity que ajuste el valor intrínseco al valor actual de la acción. Puesto que en el DDM se descuentan flujos que van a parar al accionista (no a los acreedores), una vez despejada la X del cost of equity para valor intrínseco igual a valor actual, podemos tomar ese cost of equity como el retorno anual para el accionista comprando al precio actual de la acción. Esto es lo que se llama implied cost of capital — y como decíamos, a diferencia del DCF, en un DDM ese número sí equivale al retorno total que puedo esperar como accionista (dividend yield + crecimiento), si mis proyecciones de dividendo se cumplen.

He utilizado un crecimiento esperado del dividendo a 5 años del 5%. Bastante conservador, porque si miramos el crecimiento de los últimos años, vemos que ha sido superior.

Usamos Terminal Dividend Growth Rate (g) 2.5%.

Con esta configuración, para obtener un valor intrínseco parecido al precio actual, necesitamos un cost of equity de aproximadamente 5.5%. Puesto que hemos hecho algunas estimaciones como un g de 2.5% y un crecimiento de dividendo de 5% para los próximos 5 años, es preferible tratarlo como un rango (5-6%) más que como un número con decimales de precisión. Y ese rango sería nuestra estimación de la rentabilidad que podemos esperar de McDonald's.

¿Es Eso Mucho o Poco? El Dato Incómodo

El bono del Tesoro americano a 10 años cotiza ahora mismo en torno al 4.5-4.6%. Si aplico el modelo estándar para estimar cuánto retorno debería exigir el mercado a una acción como MCD (CAPM: risk-free + beta × prima de riesgo), con una beta razonable para MCD (0.6-0.7, la que reportan la mayoría de proveedores de datos), sale un cost of equity “teórico” de en torno al 7-7.5%.

Es decir: el mercado, al precio actual, solo me está pagando un 5-6% de retorno esperado vía dividendo y crecimiento, cuando el riesgo sistemático de la acción, según el modelo estándar, justificaría exigir más, cerca del 7-7.5%. El mercado paga muy poca prima por el riesgo de McDonald’s, porque valora tanto la previsibilidad y el perfil “bono-proxy” del negocio, que le perdona parte de esa prima que otro tipo de acciones sí tendría que ofrecer.

Entonces, ¿Por Qué Sigo Comprando?

Porque ese 5-6% no es un mal retorno para el nivel de riesgo real que percibo en el negocio, aunque el CAPM de manual diría que debería exigir más. Y sobre todo porque:

Es un negocio con foso competitivo brutal, poder de fijación de precios, un componente semi-inmobiliario (los alquileres a franquiciados son básicamente una renta), y una previsibilidad de caja que muy pocas empresas del mundo tienen.

Ese retorno no depende de que el mercado me re-rate el múltiplo. Depende solo de crecimiento orgánico + dividendo, que es justamente el tipo de retorno que menos me falla cuando las cosas se ponen feas en el resto de la cartera.

Porque se ha comportado bien en momentos complicados para el mercado, por ejemplo el 2008.

Y esto es lo más importante para mí: es la posición que me deja dormir. No la compro por asimetría de rentabilidad, la compro porque a este precio el retorno esperado, aunque modesto, viene con muchísima menos incertidumbre que en casi cualquier otro sitio donde podría poner ese dinero.

Discounted Cash Flow Para Comparar con la Competencia

Voy a hacer un ejercicio parecido con un DCF clásico. Usaré un crecimiento de revenue del 3.79%, media desde 2018, y margen de conversión de revenue a FCF del 25.74%, también media desde 2018. Y un valor terminal g del 2.5%.

Y con estos valores voy a buscar un WACC con el que obtenga un valor intrínseco similar al precio actual:

Curiosamente obtengo un WACC similar al cost of equity del DDM. No podemos asumir que el WACC equivale a la rentabilidad anual para el accionista porque en el WACC va incluido lo que la empresa tiene para el accionista y para los acreedores (por ejemplo para los bonistas). Sin embargo es curioso que salga un valor similar al cost of equity del DDM. Entiendo que esto es debido a que, a pesar de que McDonald’s tiene bastante deuda, genera tanto flujo de caja, que pagar la deuda no es un problema (y que probablemente se financie bastante barato). Además hay que tener en cuenta que utilicé un crecimiento del dividendo muy conservador, de 5% solamente.

En cualquier caso mi objetivo era más obtener un WACC para el precio actual de la acción, por un lado para compararlo con el WACC consensuado por analistas para McDonald’s, que se sitúa entre 7~7.5%. Y por otro para compararlo con competidores como QSR o Starbucks.

En el caso de QSR, si realizo el mismo ejercicio, obtengo un WACC del 9% para ajustar el valor intrínseco al precio actual:

Para Starbucks obtengo un WACC del 5.3%:

Por lo tanto, si hacemos el ejercicio de buscar un WACC que fuerce el valor intrínseco al precio actual, obtenemos un WACC del 9% para QSR, del 5.5% para MCD y del 5.3% para SBUX, al precio en que se realizaron los cálculos.

Esto podría darnos una idea de que QSR podría estar más barata que MCD y SBUX. De hecho, haciendo el ejercicio que hicimos con el DDM para QSR, obtengo una rentabilidad anual del 6.3%, y para SBUX, del 5.2%.

¿Entonces, Por Qué MCD y no QSR?

McDonald’s no es solo una empresa de comida rápida, es un cartel inmobiliario. Compra o alquila los terrenos y locales más codiciados del mundo y los subalquila a sus franquiciados. Esto le asegura ingresos estables por alquiler y regalías que no dependen únicamente de la venta de hamburguesas.

QSR utiliza más un modelo puramente de franquicia, careciendo de este gigantesco portafolio de bienes raíces.

Por otro lado, gracias a su escala global y a su estructura de negocio, MCD tiene márgenes operativos líderes en el sector. MCD históricamente mantiene márgenes operativos cercanos al 46%, mientras que QSR suele rondar márgenes operativos próximos al 26%. MCD necesita generar menos crecimiento de ventas para obtener el mismo beneficio neto que sus competidores.

Además tiene una fortaleza de marca y escala superiores. McDonald’s es una de las marcas más reconocidas del mundo. Esto se traduce en un tráfico masivo y un poder de negociación con proveedores que cadenas como Burger King, Popeyes o Tim Hortons (marcas de QSR) no tienen.

Conclusión y Cuidadín

No todas las posiciones de una cartera tienen que ser la mejor idea de rentabilidad-riesgo del mundo. Algunas están para dar rentabilidad ajustada al riesgo bajo, previsibilidad, y sobre todo, paz mental para poder mantener el resto de la cartera con la cabeza fría cuando las cosas se ponen feas en otro sitio. MCD, para mí, es eso. Un ~5-6% anual, con esta calidad de negocio detrás, es un retorno por el que firmo. Y si el precio sigue cayendo, la ecuación solo mejora — sigo ampliando.

Sea como sea, como siempre, esto no es ninguna recomendación de compra. Ninguna empresa es 100% segura y mañana mismo la narrativa del mercado puede cambiar por completo con respecto a McDonald’s, como ya cambió con respecto a otros pesos pesados, como la infame Diageo. Si mañana una bacteria rarísima afecta a un montón de consumidores de McDonald’s, se va a liar. Si los chavales dejan de comer hamburguesas como supuestamente han dejado de beber o fumar (o al menos el mercado compra esa narrativa), se va a liar. Si Ozempic nos pone a todos a dieta, se va a liar. Así que como siempre, haced vuestro propio análisis y valorad los riesgos.

Un saludo a todos.

English Version

McDonald's. Returns, Peace, Tranquility.

A 5-6% annual return I can sleep soundly on.

Today I want to leave AI, oil, chips, and SaaSmageddon aside for a bit, to find a little peace and quiet 🧘 That’s exactly what I feel when I walk into a McDonald’s knowing that 4% of my portfolio is in this stock. I sit down at a table, check the app for today’s deal. And for 3~4 euros, I eat a small burger and some fries.

When I travel to another country, I often struggle to communicate because my English is terrible. And I’ll probably get lost and have no idea how to use public transport, because I have the survival instincts of a lemming.

But I know that there, I’ll find a McDonald’s where I can eat and rest for a while. A refuge. A home. And a little piece of that refuge is mine.

That’s why I keep accumulating the stock that gives me the most peace of mind in my portfolio (with all due respect to Berkshire). And even though I already own quite a bit, I’m hoping it keeps falling so I can keep buying more. Let me walk you through the numbers behind that peace of mind.

The price, In Context

MCD is currently trading at a PE ratio of around 22-23x. That’s somewhere between 10% and 13% below its 3, 5, and 10-year average, which sits around 25-26x. It’s not a historic bargain like a covid-crash moment, but it is one of the lowest multiples the stock has traded at in quite a while.

A PE of 22 for a company growing revenue at 3-4% a year isn’t cheap in absolute terms. It’s cheap relative to its own history.

Finding The Expected Return Through Cost of Equity

Since McDonald’s is a classic dividend payer, I’ll start by running a Dividend Discount Model. In this case, the idea isn’t to use the cost of equity consensus among analysts, but to force a cost of equity that makes the intrinsic value match the current share price. Since a DDM discounts cash flows that go directly to the shareholder (not to creditors), once you solve for X — the cost of equity that makes intrinsic value equal to current price — you can treat that cost of equity as the annual return for a shareholder buying at today’s price. This is what’s called implied cost of capital — and as we said, unlike a DCF, in a DDM that number really does equal the total return I can expect as a shareholder (dividend yield + growth), assuming my dividend projections hold.

I used an expected dividend growth rate of 5% over the next 5 years. Pretty conservative, since if you look at recent growth, it’s actually been higher.

We use a Terminal Dividend Growth Rate (g) of 2.5%.

With this setup, to get an intrinsic value close to the current price, we need a cost of equity of approximately 5.5%. Since we made some estimates here — a g of 2.5% and dividend growth of 5% for the next 5 years — it’s better to treat this as a range (5-6%) rather than a number with decimal-point precision. And that range would be our estimate of the return we can expect from McDonald’s.

Is That a Lot or a Little? The Uncomfortable Part

The 10-year US Treasury is currently trading around 4.5-4.6%. If I apply the standard model to estimate what return the market should demand from a stock like MCD (CAPM: risk-free rate + beta × risk premium), using a reasonable beta for MCD (0.6-0.7, what most data providers report), I get a “theoretical” cost of equity of around 7-7.5%.

In other words: at the current price, the market is only paying me a 5-6% expected return via dividend and growth, when the stock’s systematic risk, according to the standard model, would justify demanding more — closer to 7-7.5%. The market is paying very little risk premium for MCD, likely because it values the predictability and “bond-proxy” profile of the business so highly that it forgives part of the premium that other types of stocks would have to offer.

So Why do I Keep Buying?

Because that 5-6% isn’t a bad return for the actual level of risk I perceive in the business, even if textbook CAPM says I should demand more. And above all, because:

It’s a business with a brutal competitive moat, pricing power, a semi-real-estate component (the rent it charges franchisees is basically a landlord’s income), and a predictability of cash flow that very few companies in the world have.

That return doesn’t depend on the market re-rating the multiple. It depends only on organic growth + dividend, which is exactly the kind of return that fails me least when things get ugly elsewhere in the portfolio.

Because it’s held up well during rough patches for the market — 2008, for example.

And this is the most important part for me: it’s the position that lets me sleep at night. I don’t buy it for asymmetric upside, I buy it because at this price, the expected return, modest as it is, comes with far less uncertainty than almost anywhere else I could put that money.

A Discounted Cash Flow to Compare Against the Competition

I’ll run a similar exercise with a classic DCF. I’ll use a revenue growth rate of 3.79% (average since 2018) and a revenue-to-FCF conversion margin of 25.74% (also average since 2018). And a terminal g of 2.5%.

With these inputs, I’ll solve for the WACC that produces an intrinsic value close to the current price:

Interestingly, I get a WACC similar to the DDM’s cost of equity. We can’t assume that WACC equals the annual return for the shareholder, because WACC bundles together what belongs to the shareholder and what belongs to the creditors (bondholders, for example). Still, it’s curious that a similar number comes out of both. I think this is because, even though McDonald’s carries plenty of debt, it generates so much cash flow that servicing that debt isn’t a problem (and it’s probably financed pretty cheaply). It’s also worth noting I used a very conservative dividend growth assumption — just 5%.

Either way, my goal here was mostly to get a WACC for the current share price so I could compare it, on one hand, with the analyst-consensus WACC for McDonald’s, which sits between 7~7.5%. And on the other hand, to compare it against competitors like QSR and Starbucks.

For QSR, running the same exercise, I get a WACC of 9% to make intrinsic value match the current price:

For Starbucks, I get a WACC of 5.3%:

So, running the exercise of solving for the WACC that forces intrinsic value to match the current price, we get a WACC of 9% for QSR, 5.5% for MCD, and 5.3% for SBUX, at the prices these calculations were done.

This might suggest that QSR could be cheaper than both MCD and SBUX. In fact, running the same DDM exercise for QSR gives me an annual return of 6.3%, and for SBUX, 5.2%.

So Then, why MCD and not QSR?

McDonald’s isn’t just a fast-food company, it’s a real estate cartel. It buys or leases the most coveted land and locations in the world and subleases them to its franchisees. This guarantees stable rental income and royalties that don’t depend solely on burger sales.

QSR relies more on a pure franchise model, lacking this massive real estate portfolio.

On top of that, thanks to its global scale and business structure, MCD has industry-leading operating margins. MCD historically runs operating margins close to 46%, while QSR tends to hover around 26%. MCD needs less sales growth to generate the same net profit as its competitors.

It also has superior brand strength and scale. McDonald’s is one of the most recognized brands in the world. That translates into massive foot traffic and supplier bargaining power that chains like Burger King, Popeyes, or Tim Hortons (QSR’s brands) simply don’t have.

Conclusion, and a Word of Caution

Not every position in a portfolio has to be the best risk-reward idea in the world. Some exist to provide low-risk, risk-adjusted return, predictability, and above all, peace of mind — so you can keep a cool head on the rest of the portfolio when things get ugly elsewhere. For me, MCD is that. A ~5-6% annual return, backed by this quality of business, is a return I’ll happily sign up for. And if the price keeps falling, the equation only gets better — I keep adding.

Either way, as always, this is not investment advice. No company is 100% safe, and tomorrow the market narrative around McDonald’s could flip completely, just as it did with other heavyweights, like the infamous Diageo. If some rare bacteria hits a bunch of McDonald’s customers tomorrow, it’s going to get messy. If kids stop eating burgers the way they supposedly stopped drinking or smoking (or at least the way the market has bought into that story), it’s going to get messy. If Ozempic puts everyone on a diet, it’s going to get messy. So, as always, do your own research and weigh the risks.

Cheers to all.

English version below ↓

Índice.

1. Adyen, La Joven Promesa

2. Un Ecosistema En Movimiento

3. Las Viejas Glorias

4. Los Cimientos

5. ¿SaaSmageddon?

6. Las Valoraciones

7. Conclusión

En mi último post hacía una introducción al ecosistema de pagos más estandarizado a nivel mundial. Podéis verlo aquí:

https://www.poeticstreet.com/p/de-pesca-en-el-ecosistema-de-los

Poco tiempo después empecé a investigar más a fondo la empresa Adyen. Con ayuda de la IA preparé una tesis bastante completa sobre la empresa que os adjunto a continuación y que puede resultar útil para quien quiera profundizar.

Adyen Tesis

1.18MB ∙ PDF file

Download

Download

Sin embargo, creo que no tiene sentido convertir este post en la enésima tesis de inversión sobre Adyen. Ya hay grandes tesis ahí fuera que además coinciden bastante entre ellas. Los puntos fuertes de Adyen están claros, también los riesgos.

Todo el mundo parece tener relativamente claro que es una gran empresa y que, dentro del mundo de los pagos, probablemente sea la que mejor está ejecutando. Yo mismo abrí una posición recientemente en Adyen, seguramente de forma precipitada.

Desde que empecé este blog dejé claro que mi objetivo sería ser transparente con mi cartera y con mis movimientos. No hace mucho tuve Fiserv en cartera, y cerré la posición casi al momento. Y como he comentado, recientemente abrí posición en Adyen.

Sin embargo, poco después cerré media posición de Adyen también: quería asentar mi opinión sobre la empresa y ampliarla en el futuro si fuera necesario. Y hoy mismo, aún sabiendo que es una buena empresa y que hay muchas probabilidades de que lo sigan haciendo bien (y de que su precio suba en los próximos meses y años), he cerrado la posición entera.

En este post os digo por qué.

Adyen, La Joven Promesa

Antes de explicar por qué he decidido salir de la posición, quiero dejar clara una cosa.

No vendo porque piense que Adyen sea una mala empresa. De hecho, probablemente sea justo lo contrario.

Después de estudiar durante semanas el sector de los pagos y la propia empresa, mi conclusión es que Adyen es, probablemente, la mejor empresa del sector.

Probablemente tenga la mejor arquitectura tecnológica del sector. Tiene una cultura empresarial extraordinaria. Está dirigida por sus fundadores, que siguen teniendo una parte muy importante de su patrimonio invertido en la propia empresa. Asigna muy bien el capital. Su plataforma está construida sobre un único código desarrollado internamente, evitando el “Frankenstein” de adquisiciones que caracteriza a muchos competidores. Su retorno sobre el capital sigue siendo muy elevado. Su balance es prácticamente impecable. Y, además, dispone de una pista de crecimiento considerable gracias a la expansión internacional y a los nuevos productos financieros.

En otras palabras: si tuviera que elegir una única empresa dentro del sector de los pagos, probablemente elegiría Adyen.

Entonces, ¿por qué vender?

Porque una gran empresa no implica necesariamente una gran inversión. Hay una diferencia enorme entre entender un negocio y ser capaz de predecir cómo evolucionará durante la próxima década. Y ahí es donde empiezan mis dudas.

Un Ecosistema En Movimiento

El otro día fuimos a la piscina municipal del pueblo donde suelo ir algunos fines de semana. Pregunté si podía pagar con tarjeta y me dijeron que no, pero que podía pagar con un bizum. Fue una tontería pero me hizo recapacitar.

Me hizo pensar que el cliente realmente no quiere pagar con tarjeta. Ni con Bizum. Ni con Apple Pay. Ni con el método que esté de moda dentro de cinco años. El cliente simplemente quiere pagar. Y cuanto más invisible se vuelve el pago, menos claro tengo quién terminará capturando el valor económico de toda esa cadena.

Después de estudiar todo el ecosistema, me he dado cuenta de que la competencia en los pagos es mucho más intensa y dinámica de lo que inicialmente pensaba. Ya no compite únicamente contra otros procesadores. También convive con bancos tradicionales, fintechs, gigantes tecnológicos, software vertical especializado, nuevas formas de pago e incluso cambios regulatorios. Muchos de ellos cooperan entre sí, pero todos intentan capturar una parte mayor de la cadena de valor.

Y todo esto ocurre en un mercado donde la tecnología evoluciona muy deprisa y donde una ventaja competitiva puede erosionarse más rápido que en otros sectores.

Las Viejas Glorias

En el sector de pagos existen algunas piezas clave que llevan ahí desde tiempos inmemoriales. Las llamadas “legacy”.

Fiserv es el ejemplo perfecto del gigante tradicional. Opera un modelo muy diferente al de Adyen. Mientras Adyen es una plataforma única y orgánica, Fiserv es un conglomerado tecnológico-financiero construido a base de adquisiciones. Los resultados recientes de Fiserv muestran presiones en los ingresos orgánicos de sus divisiones de procesamiento y soluciones financieras tradicionales. El mercado la castiga cuando ven que el negocio viejo pesa.

Otro ejemplo es Paypal. En su momento también fue la niña bonita del mercado, pero un buen día dejó de cumplir las expectativas de crecimiento que tenía, posiblemente debido a malas adquisiciones, a una mala gestión, y sobre todo a la dura competencia (Google Pay, Apple Pay, también entran en juego aquí). El mercado castiga muchísimo más el cambio de expectativas que la pérdida de beneficios.

Sin embargo sería iluso pensar que las viejas glorias no van a luchar por defender su castillo.

Por ejemplo, Fiserv invierte en su caballo de batalla y motor de crecimiento Clover, una plataforma de punto de venta y software para pequeños y medianos negocios. Paradójicamente, aquello que hoy parece una debilidad (haber crecido mediante adquisiciones) también le proporciona una enorme base instalada sobre la que desplegar nuevos productos.

Paypal por su parte ha invertido fuertemente en Venmo, que ha dejado de ser solo una aplicación P2P para dividir gastos entre amigos, para convertirse en “Pay with Venmo”, y han acelerado la expansión de este botón de pago en miles de tiendas online, así como la tarjeta de Débito Venmo entre el público joven. Paypal también adquirió Braintree, que aunque tecnológicamente diferente, también compite con Adyen en servicios de adquirencia y en el sector del procesamiento de pagos para grandes empresas.

En resumen, pensar que viejos titanes como Fiserv, Paypal u otros no van a intentar competir y hacerse un hueco con soluciones nuevas, es pecar de iluso. Además estas grandes empresas ya están integradas en el ecosistema de pagos y puede implantar con más facilidad sus nuevas soluciones.

Los Cimientos

Pienso en la base, en los cimientos de todo el ecosistema de pagos.

Todo el mundo da por sentado que empresas por ejemplo como Visa o Mastercard tienen un moat gigantesco, que el mundo se paraliza sin ellas. Y es verdad.

Pero estas empresas aparecieron para resolver un problema de infraestructura que existía hace treinta años, cuando Internet no era lo que es hoy y la capacidad de procesamiento no era la misma. Y sin embargo se requerían respuestas inmediatas cuando se realizaba un pago en un comercio.

La pregunta que me hago es si seguirá siendo necesario dentro de diez o veinte años que toda la infraestructura de pagos se realice en una VPN con líneas dedicadas propias ultrarápidas, segregada del internet público, sobre un protocolo antiguo.

Quiero decir, el Internet de hoy es veloz. Los servidores en la nube son rápidos, pueden procesar capas de cifrado sin apenas introducir retardos. Esto es solo una reflexión mía. Puede que esté completamente equivocado. Pero honestamente, creo que hoy en día se pueden construir perfectamente sistemas de pago alternativos sobre internet, sin la necesidad de canales dedicados ultra costosos. Utilizando algoritmos y protocolos modernos.

Además, no es necesario reemplazar de golpe redes como la de Visa y Mastercard o cualquier otro sistema heredado. Pueden coexistir soluciones modernas con las soluciones legacy.

A lo que voy es a lo mismo que ya he mencionado. Que el ecosistema de pagos está en una fase donde todo puede cambiar y donde hay muchos competidores: los que están desde siempre, los que traen nuevas soluciones, y los que quizás todavía no existen. Todos ellos diseñando soluciones nuevas. Porque la infraestructura actual lo permite.

Y si necesitáis una prueba de que esto no es una elucubración mía, la tenéis fresquísima, de esta misma semana. Ayer mismo se supo que JPMorgan Chase, Bank of America, Wells Fargo y PNC han mantenido conversaciones preliminares para comprarle a Fiserv su red de débito, la STAR Network. El motivo es exactamente el que imaginaréis: saltarse el duopolio de Visa y Mastercard y esquivar los límites federales a las comisiones que impone la Enmienda Durbin del Dodd-Frank. Al poseer su propia red, los bancos recuperarían miles de millones en comisiones que hoy se dejan por el camino.

El ecosistema de pagos se está fragmentando desde sus cimientos y los propios creadores del sistema están buscando formas de hackear los rieles tradicionales sobre los que se ha edificado la industria durante las últimas décadas.

Y si los propios cimientos del sistema están en juego, ¿cómo podemos imaginar de qué forma van a cambiar las cosas en los próximos diez o veinte años, y qué empresas van a estar ahí para llevarse el gato al agua?

¿SaaSmageddon?

Contra Adyen se podría esgrimir un argumento parecido al que ya se usa contra otras compañías de software: si la ventaja competitiva reside en un código limpio, unificado y construido desde cero, entonces la IA se convierte en un arma de doble filo. Ayuda a que los competidores con software peor, heredado y lleno de parches, puedan ponerse las pilas más rápido de lo habitual. Y eso, en teoría, erosiona parte del foso de Adyen.

Sin embargo, no creo que sea un argumento con demasiado peso hoy por hoy. De momento la IA está jugando más a favor de Adyen que en su contra: la usan para pelear contra el fraude y para elegir las mejores rutas de enrutamiento en cada transacción.

Solo quería dejar constancia de este argumento, que ya me he encontrado en otras ocasiones.

Las Valoraciones

Hemos hablado de jóvenes promesas como Adyen y de viejas glorias como Fiserv o Paypal. Y hemos dejado clara la ventaja de Adyen sobre las mismas, sin olvidar que los jugadores de siempre también harán lo suyo por comerse un trozo del pastel y por hacerse un hueco con sus nuevas soluciones.

Pero aquí hay algo muy importante de lo que no hemos hablado. La valoración.

Adyen cotiza a un PER de casi 26 en el momento de escribir estas líneas. Mientras Paypal o Fiserv suelen moverse en entornos de PER 8~9.

Aunque Adyen es la empresa posiblemente más prometedora del sector de pagos, se espera mucho de ella. Y precisamente ahí reside el problema. Cuando el mercado exige perfección, basta con hacerlo “muy bien” en lugar de “extraordinariamente bien” para que la valoración se resienta.

Mientras tanto Fiserv o Paypal ya han sufrido el castigo por el traspiés, y seguramente un castigo excesivo para dos empresas que siguen produciendo FCF a raudales.

Si hacemos un DCF con el crecimiento medio del revenue desde 2018 de 11.68 y conversión de revenue a FCF media de 20.51, para un WACC de 25% y tasa de crecimiento perpetuo de 2.5%, obtenemos un valor intrínseco por DCF de 45.73 dólares para Paypal:

Y si hacemos un ejercicio parecido para Fiserv, con crecimiento revenue de 5.37%, conversión a FCF de 21.6%, WACC 11.5% y g 2.5%, obtenemos un valor intrínseco de 55.72 dólares:

Por supuesto habría que estimar otros ratios de crecimiento y de conversión a FCF para escenarios futuros más lúgubres, solamente es para ejemplificar cómo es el mercado. En el momento que huele problemas o las expectativas de crecimiento cambian, el castigo es implacable.

Adyen está haciendo las cosas bien y mejor que sus competidores, y aunque ya lleva caída desde los múltiplos descabellados que tenía, cualquier error cometido por ellos mismos, o cualquier cambio en el ecosistema de pagos que, aunque ajeno a ellos, les afecte, puede tirar su valoración a la baja aún más.

En mi humilde opinión, en el sector de los pagos hay empresas ya castigadas, que en momentos de castigo excesivo, podrían dar lugar a un trade de una duración determinada. Pero no me parece un sector para comprar y olvidar a largo plazo. Ahora mismo me planteo más intentar cazar rebotes en empresas a PER 8 como Fiserv o Paypal que entrar en Adyen a largo plazo.

Conclusión

Cuando empecé a estudiar Adyen, buscaba confirmar que era una empresa excepcional. Y terminé convencido de que lo es. Lo que no esperaba era que el análisis me llevara a cuestionar el propio sector. Cuanto más entendía Adyen, más entendía que el éxito futuro no dependerá sólo de ejecutar mejor que sus competidores, sino también de cómo evolucione una industria que todavía está redefiniéndose. Esa fue la conclusión que me hizo cambiar de opinión.

En este post no pretendo decir que Adyen no vaya a tener un buen futuro, porque la verdad que lo más probable es que lo tenga.

Lo que digo es que no soy capaz de estimar con suficiente confianza la probabilidad de que ese buen futuro ocurra. Y esa diferencia es muy importante.

Todas las inversiones son probabilísticas y por tanto apuestas. La diferencia es que hay negocios en los que siento que puedo estimar esas probabilidades con bastante confianza, y otros en los que no. No necesito encontrar únicamente empresas excelentes; necesito encontrar aquellas cuyo futuro pueda imaginar con un grado razonable de confianza.

Invertir en algo con una estimación de 50~60% de probabilidad, suena más a apostar en el casino. Pero si tu convicción es superior, si te sientes parte de la empresa porque te gusta lo que hacen y porque te sientes a gusto en ese barco, si estás deseando que baje el precio para ampliar posición, en ese caso puedes invertir a diez o quince años estando tranquilo, y eso sí es invertir de verdad.

En consumo básico, o en ciertas empresas de software empresarial, en infraestructuras o incluso determinados negocios industriales siento que puedo hacerlo, que puedo confiar. En pagos, hoy por hoy, no. Porque las aguas del mundo de los pagos están demasiado revueltas, incluso para buenos pescadores.

Puede que dentro de diez años Adyen sea mucho más grande que ahora. También puede que aparezcan nuevos competidores, nuevos estándares tecnológicos o nuevos modelos de negocio que cambien completamente el equilibrio del sector.

Quizá dentro de cinco años vuelva a comprar Adyen. O quizá no. Lo importante no es acertar con una empresa concreta. Lo importante es invertir únicamente cuando siento que estoy navegando en aguas que realmente entiendo.

Como siempre, nada de esto es consejo de inversión. Ademas puedo estar totalmente equivocado, de hecho es lo más probable.

Si os ha gustado el artículo, suscribíos para que os lleguen más a vuestro correo electrónico. Cualquier comentario o corrección es bienvenido.

¡Un fuerte abrazo!

English Version

I Fell in Love With a Company and Broke Up With an Entire Sector

A lesson on the limits of your circle of competence. Why I sold the best company in a sector I can't manage to predict.

In my last post I wrote an introduction to the world’s most standardized payments ecosystem. You can check it out here:

https://www.poeticstreet.com/p/de-pesca-en-el-ecosistema-de-los

Shortly after, I started digging deeper into Adyen. With some help from AI I put together a fairly thorough thesis on the company.

That said, I don’t think it makes sense to turn this post into yet another Adyen investment thesis. There are already great ones out there, and most of them land in pretty similar places. Adyen’s strengths are clear, and so are its risks.

Pretty much everyone agrees it’s a great company, and that within the payments world it’s probably the one executing best. I myself recently opened a position in Adyen, likely a bit hastily.

Since I started this blog I’ve made it clear that one of my goals is to be transparent about my portfolio and my moves. I bought Fiserv, and closed that position the next day. And as I mentioned, I recently opened a position in Adyen.

Shortly after, though, I closed half the Adyen position too: I wanted to firm up my view on the company before potentially adding to it later. And today, even knowing it’s a good business with a decent chance of continuing to execute well (and of its price climbing over the coming months and years), I’ve closed the position entirely.

In this post I’ll tell you why.

Adyen, The Young Prodigy

Before explaining why I decided to exit, I want to make one thing clear.

I’m not selling because I think Adyen is a bad company. If anything, it’s probably the opposite.

After weeks of studying the payments sector and the company itself, my conclusion is that Adyen is, probably, the best company in the sector.

It probably has the best tech architecture in the sector. It has an extraordinary corporate culture. It’s run by its founders, who still have a very significant part of their net worth invested in the company. It allocates capital very well. Its platform is built on a single codebase developed in-house, avoiding the “Frankenstein” of bolted-on acquisitions that characterizes many competitors. Its return on capital remains very high. Its balance sheet is practically spotless. And on top of that, it has a considerable runway for growth thanks to international expansion and new financial products.

In other words: if I had to pick a single company within the payments sector, I’d probably pick Adyen.

So... why sell?

Because a great company doesn’t necessarily make a great investment. There’s a huge difference between understanding a business and being able to predict how it will evolve over the next decade. And that’s where my doubts start.

An Ecosystem In Motion

The other day we went to the local municipal pool in the town where I spend some weekends. I asked if I could pay by card and was told no, but I could pay with Bizum. It was a silly little moment, but it made me stop and think.

It made me realize that the customer doesn’t actually want to pay with a card. Or with Bizum. Or with Apple Pay. Or with whatever method happens to be trendy five years from now. The customer just wants to pay. And the more invisible the payment itself becomes, the less clear it is to me who ends up capturing the economic value along that whole chain.

After studying the entire ecosystem, I’ve realized that competition in payments is much more intense and dynamic than I initially thought. It’s no longer just competing against other processors. It also coexists with traditional banks, fintechs, tech giants, specialized vertical software, new payment methods, and even regulatory shifts. Many of these players cooperate with each other, but all of them are trying to capture a bigger slice of the value chain.

And all of this is happening in a market where technology moves fast, and where a competitive edge can erode faster than in other sectors.

The Old Guard

The payments sector has a few key pieces that have been around since time immemorial. The so-called “legacy” players.

Fiserv is the perfect example of the traditional giant. It runs a very different model from Adyen’s. While Adyen is a single, organically-built platform, Fiserv is a tech-financial conglomerate built through acquisitions. Fiserv’s recent results show pressure on organic revenue in its processing and traditional financial solutions divisions. The market punishes it whenever the old business shows its weight.

Another example is PayPal. It too was once the market’s golden child, but at some point it stopped meeting the growth expectations set for it, likely due to bad acquisitions, poor management, and above all fierce competition (Google Pay and Apple Pay play a role here too). The market punishes a shift in expectations far more harshly than it punishes an actual drop in profits.

Still, it would be naive to think the old guard won’t fight to defend its castle.

Fiserv, for instance, is pouring money into its growth engine and workhorse, Clover, a point-of-sale and software platform for small and medium businesses. Paradoxically, what looks like a weakness today (having grown through acquisitions) also gives it a massive installed base on which to deploy new products.

PayPal, for its part, has invested heavily in Venmo, which has stopped being just a P2P app for splitting bills among friends and has turned into “Pay with Venmo,” aggressively expanding this payment button across thousands of online stores, along with the Venmo debit card aimed at younger users. PayPal also acquired Braintree, which, while technologically different, also competes with Adyen in acquiring services and enterprise payment processing.

In short, assuming that old titans like Fiserv, PayPal, and others won’t try to compete and carve out room with new solutions would be naive. On top of that, these large companies are already embedded in the payments ecosystem and can roll out new solutions more easily.

The Foundations

I like to think about the base, the actual foundations, of the whole payments ecosystem.

Everyone assumes that companies like Visa or Mastercard have a gigantic moat, that the world would grind to a halt without them. And that’s true.

But these companies emerged to solve an infrastructure problem that existed thirty years ago, when the internet wasn’t what it is today and processing power wasn’t the same. Even so, instant responses were required whenever a payment was made at a merchant.

The question I keep asking myself is whether, ten or twenty years from now, it will still be necessary to run all of payments infrastructure over a VPN with dedicated, ultra-fast private lines, segregated from the public internet, running on an old protocol.

I mean, today’s internet is fast. Cloud servers are fast, capable of handling encryption layers with barely any added latency. This is just my own reflection. I could be completely wrong. But honestly, I think it’s entirely possible today to build alternative payment systems over the public internet, without needing ultra-expensive dedicated channels, using modern algorithms and protocols.

Besides, there’s no need to replace networks like Visa’s and Mastercard’s, or any other legacy system, overnight. Modern solutions can coexist with legacy ones.

What I’m getting at is the same point I’ve already made. The payments ecosystem is in a phase where everything can change, and where there are many competitors: the ones that have always been there, the ones bringing new solutions, and the ones that maybe don’t even exist yet. All of them designing new solutions. Because today’s infrastructure allows it.

And if you need proof that this isn’t just something I made up, it’s about as fresh as it gets, from this very week. Just yesterday it came out that JPMorgan Chase, Bank of America, Wells Fargo, and PNC have held preliminary talks about buying Fiserv’s debit network, the STAR Network. The reason is exactly what you’d expect: to bypass the Visa/Mastercard duopoly and dodge the fee caps imposed by the Durbin Amendment. By owning their own network, banks would recover billions in fees they currently leave on the table.

The payments ecosystem is fragmenting from its very foundations, and the system’s own creators are looking for ways to hack the traditional rails the industry has been built on for decades.

And if the very foundations of the system are in play, how can we possibly picture how things will change over the next ten or twenty years, and which companies will be there to actually cash in?

SaaSmageddon?

A similar argument to the one already used against other software companies could be made against Adyen: if its competitive edge lies in clean, unified code built from scratch, then AI becomes a double-edged sword. It helps competitors running worse, legacy, patchwork software catch up faster than they normally would. And in theory, that erodes part of Adyen’s moat.

That said, I don’t think this argument carries much weight today. Right now AI is working more in Adyen’s favor than against it: they use it to fight fraud and to pick the best routing paths for each transaction.

I just wanted to flag this argument, since I’ve come across it a few times before.

Valuations

We’ve talked about young prodigies like Adyen and old guard names like Fiserv or PayPal. We’ve made clear Adyen’s edge over them, without forgetting that the incumbents will also fight to grab a bigger slice of the pie and carve out room for their own new solutions.

But there’s something important we haven’t touched on yet. Valuation.

Adyen trades at a P/E of nearly 26 as I write this. Meanwhile PayPal and Fiserv tend to trade in the 8-9x P/E range.

Even though Adyen is arguably the most promising company in the payments sector, a lot is expected of it. And that’s exactly where the problem lies. When the market demands perfection, doing “very well” instead of “extraordinarily well” is enough to hurt the valuation.

Meanwhile, Fiserv and PayPal have already taken their punishment for stumbling, likely an excessive one for two companies still generating FCF hand over fist.

If we run a DCF using the average revenue growth since 2018 of 11.68% and an average revenue-to-FCF conversion of 20.51%, with a WACC of 25% and a perpetual growth rate of 2.5%, we get an intrinsic DCF value of $45.73 for PayPal:

And running a similar exercise for Fiserv, with revenue growth of 5.37%, FCF conversion of 21.6%, a WACC of 11.5%, and g of 2.5%, we get an intrinsic value of $55.72:

Of course, one should model other growth and FCF-conversion assumptions for gloomier future scenarios; this is just to illustrate how the market behaves. The moment it smells trouble, or growth expectations shift, the punishment is merciless.

Adyen is doing things right, and better than its competitors, and even though it’s already come down a long way from the once-absurd multiples it commanded, any misstep on its own part, or any shift in the payments ecosystem that, even if unrelated to Adyen, still affects it, could push its valuation even lower.

In my humble opinion, the payments sector has names that are already beaten down, and in moments of excessive punishment, could set up a trade with a defined time horizon. But it doesn’t strike me as a sector to buy and hold forever. Right now I’m more inclined to hunt for bounces in P/E-8 names like Fiserv or PayPal than to hold Adyen for the long run.

Conclusion

When I started studying Adyen, I was looking to confirm it was an exceptional company. And I ended up convinced that it is. What I didn’t expect was for the analysis to lead me to question the sector itself. The more I understood Adyen, the more I understood that its future success won’t depend solely on out-executing its competitors, but also on how an industry that’s still redefining itself evolves. That was the conclusion that changed my mind.

With this post I’m not trying to say Adyen won’t have a good future, because honestly, it most likely will.

What I’m saying is that I can’t estimate, with enough confidence, the probability that this good future actually plays out. And that distinction matters a great deal.

All investing is, at bottom, probabilistic, and therefore a bet. The difference is that there are businesses where I feel I can estimate those probabilities with real confidence, and others where I can’t. I don’t just need to find excellent companies; I need to find ones whose future I can picture with a reasonable degree of confidence.

Investing in something with a 50-60% probability estimate feels more like a bet at the casino. But if your conviction runs higher, if you feel part of the company because you like what it does and feel comfortable being on that ship, if you’re actually hoping the price drops so you can add to your position, then you can invest with a ten- or fifteen-year horizon and sleep just fine. That’s what real investing looks like.

In consumer staples, certain enterprise software companies, infrastructure, or even some industrial businesses, I feel I can do that, that I can trust it. In payments, right now, I can’t. Because the waters of the payments world are too choppy, even for good fishermen.

Ten years from now, Adyen might be a much bigger company than it is today. New competitors, new technological standards, or new business models might also emerge and completely reshuffle the sector’s balance.

Maybe in five years I’ll buy Adyen again. Or maybe I won’t. What matters isn’t getting one specific company right. What matters is only investing when I feel I’m sailing in waters I truly understand.

As always, none of this is investment advice. I could also be completely wrong, and honestly, that’s the most likely outcome.

If you enjoyed the article, subscribe to get more of these straight to your inbox. Any comments or corrections are always welcome.

A big hug!

Se dice que a río revuelto, ganancia de pescadores. Y si te has dado una vuelta por redes sociales o por las noticias sobre economía e inversión, ya te habrás dado cuenta de que todo el ecosistema de los pagos está bastante revuelto:

• Competencia feroz entre actores tradicionales y nuevos entrantes.

• Disrupción tecnológica constante.

• Criptomonedas y stablecoins empujando desde fuera del sistema.

• Bancos centrales liándola parda con sus propias iniciativas digitales.

• Regulación y control fiscal apretando cada vez más.

• Pagos A2A (account to account), ofrecidos por algunos adquirentes para pagar directamente con el banco, de una cuenta a otra, saltándose gran parte del ecosistema tradicional, dejando fuera, por ejemplo, a Visa y Mastercard.

• Pagos Closed-Loop de saldo a saldo dentro de la misma aplicación.

• Inteligencia artificial rediseñando procesos que llevaban décadas sin tocarse.

• Y hasta el “tap to pay“, que convierte el móvil en un datáfono y deja fuera aparatos que antes eran imprescindibles.

Es un mundo que yo diría que es sencillo en el fondo, porque todo tiene su lógica cuando lo entiendes. Pero en el que ahora mismo está todo un poco liado. Algo que en principio parece simple (”doy la orden para que el dinero de la fuente X, normalmente una cuenta bancaria, aparezca en el destino Y, también una cuenta bancaria habitualmente”) se ha convertido, con décadas de desarrollo por parte de distintos actores, en un Frankenstein con muchas capas de funcionalidad encima.

Cuando ves todo el ecosistema junto, la sensación que da es que podría simplificarse enormemente. Y ese es, en mi opinión, el mayor peligro estructural para muchas de las empresas del sector. Con las redes y las tecnologías que había hace dos décadas, todo tenía sentido. Pero con la tecnología que existe hoy, servidores tremendamente potentes y un mundo 100% interconectado con redes veloces, mucha de la funcionalidad parece redundante — capas que podrían eliminarse o colapsarse, dejando fuera a actores que hoy cobran por estar en medio.

En próximos posts voy a analizar varias empresas de este sector, y este artículo es una introducción necesaria para entender el ecosistema de los pagos de forma general. En las próximas secciones intentaré dar una visión global del mismo.

⚠️ Aviso a navegantes: se viene ladrillazo ⚠️

Las próximas secciones podrían ser un poco espesas, pero creedme que merece la pena tener un poco de paciencia, porque una vez tienes el contexto y entiendes todo el ecosistema de pagos, es mucho más fácil entender el porqué de cada empresa, qué hace cada una, cómo de importante es y por qué tiene moat o no, o por qué peligra su existencia.

Movimientos B2B: Banco-Banco

La base de todo este ecosistema es la transferencia de capital entre bancos, desde una cuenta bancaria hasta otra. El resto de capas (toda la red para pagos con tarjeta de Visa o Mastercard, los adquirentes como Adyen o Fiserv, los pagos A2A, etc…) al final podrían considerarse complementos para facilitar y hacer más cómodas y seguras estas transferencias entre cuentas bancarias.

“Transferencia entre cuentas bancarias” suena a algo sencillo, y en realidad debajo hay una infraestructura brutal que poca gente conoce.

Lo primero que hay que entender es que los bancos no se comunican por el mismo internet que usas para ver Netflix. Los grandes bancos (los llamados Tier 1: JPMorgan, Citigroup, BNP Paribas, Bank Of America, Santander...) y los bancos centrales usan lo que se conoce como fibra oscura: filamentos de fibra óptica alquilados a largo plazo a operadoras de telecomunicaciones que están dedicados exclusivamente a ese tráfico. Nadie más pasa por ahí. Es literalmente un tubo de luz privado del punto A al punto B. De esta manera consiguen aislar la red y evitar en gran medida problemas de ciberseguridad.

Para conexiones internacionales entre las instituciones financieras más importantes, la historia es parecida pero a mayor escala. SWIFT (la cooperativa que conecta a más de 11000 instituciones financieras de todo el mundo) no se ha dedicado a tirar sus propios cables cruzando el Atlántico. Sin embargo sí opera su propia red privada global (SWIFTNet, una VPN de escala gigante basada en tecnología MPLS), y para ello utiliza la infraestructura física de las grandes telecos, pero de una forma completamente segregada del internet público: por así decirlo, utiliza el mismo canal físico (los mismos “cables gordos” de internet) pero tiene canales lógicos propios, “paralelos” a internet.

Además, algunas zonas pueden tener sus propias “carreteras” internas. Por ejemplo en Europa existe SEPA (Single Euro Payments Area) y SEPA Instant (la versión Fórmula 1 de SEPA). Para transferencias dentro de Europa, no llega a usarse SWIFT. En España específicamente existe Iberpay (SNCE).

Los bancos más pequeños, los que no pueden permitirse usar este tipo de infraestructura, sí usan el internet convencional, pero con cifrado de grado militar.

La banca corresponsal

Aquí llega un concepto importante, que es el que realmente explica cómo funciona el sistema: la banca corresponsal. Y algo importante para entender el ecosistema: SWIFT no mueve dinero. Mueve mensajes. Es básicamente un sistema de mensajería ultrasegura entre bancos. El dinero en sí se mueve a través de otro conjunto de actores: las cámaras de compensación.

• Mensajería interbancaria.

• SWIFT.

• Liquidación y compensación (el movimiento real del dinero).

• En Europa: TARGET2 (operado por el Eurosistema, el banco central) y EBA Clearing.

• En EE.UU: FedNow (el sistema de la Reserva Federal, relativamente reciente) y The Clearing House (opera CHIPS y RTP).

Estos sistemas de liquidación son lo que se conoce como RTGS (Real-Time Gross Settlement), y son operados mayoritariamente por los propios bancos centrales. Son la fontanería real del sistema financiero.

Los grandes bancos internacionales tienen cuentas abiertas unos en otros. El Santander tiene una cuenta con dinero real depositado en JPMorgan (y viceversa). Cuando el Santander quiere mandar dólares a JPMorgan en Nueva York, SWIFT es el canal por el que viaja la instrucción segura: “Oye, JPMorgan, soy Santander. Quítame 1000$ de mi cuenta contigo y dáselos al cliente con cuenta Y”. Cuando el Santander manda esa instrucción por SWIFT, JPMorgan simplemente hace un apunte contable: descuenta el importe de la cuenta que el Santander tiene con ellos y lo abona al destinatario final. No hay movimiento físico de dinero en ese momento, solo un ajuste de saldos.

Y si el banco que quiere enviar no tiene una relación directa con el banco destino, utiliza uno o varios bancos intermediarios que sí la tengan. Eso es lo que hace que algunas transferencias internacionales sean lentas (1-3 días hábiles) y caras: hay que pagar a cada intermediario que toca la cadena.

Ahora bien, si el Santander hace muchas transferencias hacia EE.UU y pocas en sentido contrario, esa cuenta que tiene en JPMorgan se va vaciando. En algún momento hay que reponerla con dinero real, y eso se hace mediante liquidaciones reales entre bancos a través de los sistemas de los bancos centrales: los sistemas RTGS ya comentados, como TARGET2 en Europa o Fedwire en EE.UU. Ahí sí se mueve dinero de banco central a banco central, que es el dinero “real” en el fondo del sistema.

Las limitaciones del sistema Banco-Banco

El sistema que acabamos de describir funciona. Y lleva décadas funcionando. Pero tiene fricciones evidentes: es caro en pagos internacionales, no opera 24/7, y en cuanto sales de tu zona geográfica aparecen intermediarios, tiempos de espera y comisiones que se acumulan. Además, todo este ecosistema está pensado para mover cantidades grandes entre instituciones, no para que tú pagues un café o compres algo por internet.

Y ahí está precisamente el origen de la siguiente capa del ecosistema. Visa y Mastercard no nacieron para competir con SWIFT ni con las cámaras de compensación. Nacieron para resolver un problema completamente distinto: cómo hacer que un particular pueda pagar en cualquier comercio del mundo de forma cómoda, rápida y segura, sin tener que mover dinero entre cuentas bancarias cada vez que compra algo.

Pagos C2B: Del Consumidor al Negocio

En un pago convencional con tarjeta, el flujo lógico es:

Comercio → Gateway → Adquirente → Redes de Pago (Visa/MC) → Banco Emisor

Y el dinero hace exactamente el camino inverso durante la fase de liquidación, por las noches, a través de las cámaras de compensación que ya vimos antes.

Las redes de pago como Visa o Mastercard operan sobre infraestructura privada — básicamente una VPN enorme con contratos millonarios con las operadoras de telecomunicaciones, completamente segregada del internet público. Un retardo de unos pocos milisegundos más allá de lo acordado deriva en indemnizaciones importantes. Lo que ofrecen, en esencia, es velocidad en la autorización del pago y la infraestructura para que después todo cuadre.

La magia del mensaje en 200 milisegundos

Redes como VisaNet o Banknet son capaces de procesar miles de transacciones por segundo con una latencia casi nula. Cuando el cliente pasa la tarjeta por el datáfono, ocurren dos fases que operan a velocidades y en mundos totalmente distintos:

Fase 1: Autorización (tiempo real, menos de 1 segundo)

Un mensaje viaja siguiendo el flujo de software mencionado desde el comercio hasta el banco emisor. El banco dice “este cliente tiene saldo o crédito suficiente, aprobado“. Ese OK bloquea el dinero en la cuenta del cliente y le asegura al adquirente — y por tanto al comercio — que el pago va a producirse. Nadie ha movido un solo céntimo todavía, pero el café ya es tuyo.

Fase 2: Compensación y Liquidación (Clearing & Settlement — diferido, por la noche)

Aquí es donde el “Mundo Tarjeta” se conecta con el “Mundo Transferencia”.

A lo largo del día se van acumulando autorizaciones en cada pago. Al final del día, cuando todos los comercios cierran sus terminales, los adquirentes vuelven a hablar con Visa, Mastercard o la red de pagos correspondiente. Empaquetan millones de transacciones y las mandan para hacer el “neteo”: la compensación entre todos los bancos. Y a través de los sistemas RTGS de los bancos centrales que ya comentamos (TARGET2, FedWire), se realiza el movimiento real del dinero.

Es decir: primero se autorizan y “anticipan” todos los pagos a lo largo del día, y cada cierto tiempo se hace recuento de todo el dinero que realmente tiene que moverse. Las redes de pago hacen la magia de esa compensación.

“A ver, Santander: tus clientes han gastado 10M fuera de tu banco. Pero a los comercios que tienen cuenta contigo les han comprado 2M desde tarjetas de otros bancos. En el neto global del día, estás en negativo por 8 millones. Tienes que pagarlos.”

Y el neteo consiste precisamente en eso: el Santander no hace miles de transferencias para pagar los 10M y recibir los 2M. Simplemente salda su posición neta diaria — los 8 millones — a través del sistema RTGS con el que opera su banco central.

¿Y para qué sirven los adquirentes?

Tras leer las secciones anteriores podría parecer que los adquirentes son solo un intermediario entre el comercio y la red de pagos que chupa del bote. Y efectivamente son un intermediario chupando del bote, pero en las siguientes líneas veremos que realmente se ganan el pan.

Uno podría pensar: si la red de pagos ya hace todo esto, ¿por qué el datáfono no le da los datos del cliente directamente a Visa o a Mastercard para que ellos autoricen y más tarde liquiden el pago? ¿Por qué Apple, Spotify o Uber pagan millones a Adyen en lugar de conectarse directamente a Visa?

La razón principal es que el mundo es caótico. Hay millones de comercios, cada uno con sus particularidades — su zona geográfica, su tipo de negocio, su cliente.

Conectar ese caos con las ultra-estrictas redes de Visa y Mastercard es una pesadilla de ingeniería, regulación, riesgo de fraude y gestión de tesorería. Es como intentar enchufar una tostadora doméstica directamente a una torre de alta tensión. Necesitas un transformador que adapte la potencia, gestione los picos de tensión, te asegure que no salga ardiendo la casa y, además, te agrupe la factura a fin de mes. Ese transformador es el adquirente.

Traductor

El adquirente hace de traductor: toma el formato web de Uber o el código del datáfono de la esquina, lo limpia, lo cifra y se lo entrega a la red de pagos masticado y perfecto. Si la red de pagos tiene un problema en un momento dado, el adquirente retiene la petición y la reintenta, salvando la venta del comercio.

Smart Routing

Si te conectaras directamente a Visa, estarías atrapado en su camino único. Un adquirente global como Adyen hace Smart Routing: cuando metes tu tarjeta en una web internacional, el adquirente analiza en milisegundos de qué banco es la tarjeta, de qué país, qué red es más barata en ese momento. Si ve que procesar el pago por Visa internacional va a costar un 2% y tiene una probabilidad de rechazo del 15% por sospecha de fraude, puede decidir enrutar ese pago a través de una red local o un acuerdo directo con el banco emisor. Esto maximiza la tasa de conversión — que el pago no sea denegado por error —, que es lo que realmente le importa a un comercio online.

Escudo Anti-Fraude

El adquirente asume el riesgo de crédito y fraude del comercio. Por eso, para ser adquirente necesitas una licencia financiera. Adyen analiza la salud financiera de Spotify o de la aerolínea de turno antes de contratarlos. Si la aerolínea quiebra y miles de clientes reclaman sus vuelos, Adyen tiene que devolver ese dinero de su propio bolsillo. Visa no quiere asumir el riesgo de lidiar con millones de comercios; prefiere exigirle responsabilidades a un puñado de adquirentes regulados y que ellos se las arreglen con cada cliente y cada comercio.

Tesorería

Imagina que un comercio vende mil cafés al día a 2€ cada uno. Si Visa liquidara directamente con el comercio, tendría que hacer mil micro-transferencias diarias — ineficiente y carísimo. El adquirente actúa como embudo: recibe de los bancos emisores un único bloque de fondos netos correspondiente a millones de transacciones, descuenta su comisión (más la de Visa y la del banco emisor), y le ingresa al comercio un único pago limpio al día o a la semana. El comercio no concilia nada; solo ve un ingreso diario en su cuenta.

Pagos U2U: Usuario-Usuario (P2P Peer-to-Peer)

El tercer gran bloque del ecosistema es el más cercano al usuario de a pie: mandarte dinero a un amigo para pagar la cena, dividir el alquiler, o saldar una deuda. En teoría es el caso de uso más sencillo de todos — dos personas, dos cuentas, una cantidad. En la práctica, durante décadas fue también el más incómodo, porque exigía conocer el IBAN del otro, esperar un día hábil y rezar para no equivocarse de número.

Lo que ha cambiado en los últimos años es que estos pagos ya no necesitan pasar por las redes de tarjetas. Usan directamente otros raíles bancarios que ya describimos — SEPA Instant, sistemas RTGS, APIs de Open Banking — pero con una capa de software encima que hace que el proceso sea tan rápido y sencillo como mandar un mensaje de WhatsApp.

Los ejemplos más relevantes a nivel global:

Bizum (España): Corre sobre Iberpay y SEPA Instant. En lugar de pedir el IBAN, usas el número de móvil. El dinero llega en segundos.

Zelle (EE.UU.): Consorcio creado por los grandes bancos americanos para competir con las apps de terceros. Opera directamente entre cuentas bancarias, sin saldo intermediario.

Venmo / Cash App (EE.UU.): Modelo diferente — el dinero queda en un saldo dentro de la propia app (lo que se llama “closed-loop” o bucle cerrado) hasta que el usuario decide transferirlo a su banco. Más flexible para el usuario, pero el dinero no está en tu cuenta bancaria real hasta que lo sacas.

Pix (Brasil): Creado directamente por el Banco Central brasileño. Uno de los casos de éxito más llamativos a nivel global — en pocos años se convirtió en el método de pago dominante en Brasil, desbancando tanto a las transferencias tradicionales como al efectivo.

UPI (India): La Unified Payments Interface india es probablemente el ecosistema P2P más grande del mundo por volumen de transacciones. Un caso de libro de cómo la intervención regulatoria puede acelerar la adopción masiva de pagos digitales.

Lo interesante de esta capa, desde el punto de vista inversor, es que es donde más claramente se ve la amenaza para el ecosistema tradicional. Estos sistemas no pagan tasas de intercambio a Visa ni a Mastercard. No necesitan adquirentes. Van directos de cuenta a cuenta. Y en los países donde han despegado de verdad — Brasil, India, China con WeChat Pay y Alipay — la penetración de las redes de tarjetas tradicionales ha quedado estructuralmente limitada.

Conclusión: El mapa antes del territorio

Si has llegado hasta aquí, enhorabuena — y gracias por la paciencia. Era un ladrillazo, lo sé. Pero era un ladrillazo necesario.

El ecosistema de pagos es de esos sectores donde es muy fácil leer titulares (”Visa gana X millones”, “PayPal pierde cuota de mercado”, “las stablecoins van a cargarse a los bancos”) sin tener claro exactamente qué está pasando por debajo. Y sin ese contexto, es muy difícil saber si el titular importa, si el riesgo es real, o si es simplemente ruido.

Lo que hemos visto en este artículo es que el ecosistema tiene tres grandes capas bien diferenciadas. La capa B2B, que es la fontanería real del sistema financiero — SWIFT, corresponsales, cámaras de compensación — y que lleva décadas funcionando pero arrastra fricciones evidentes en los pagos internacionales. La capa C2B, donde Visa y Mastercard construyeron su imperio precisamente para resolver lo que el sistema B2B no podía hacer: que un consumidor pague en cualquier comercio del mundo de forma cómoda y segura. Y la capa P2P, que es la más nueva y la que más claramente amenaza a las capas anteriores, porque va directa de cuenta a cuenta sin necesitar ni redes de tarjetas ni adquirentes.

Y encima de todo esto: disrupción por todos los flancos. Pagos A2A, stablecoins, Open Banking, iniciativas de bancos centrales. Todos atacando alguna de estas capas. Todos con argumentos reales. Ninguno, de momento, con una victoria decisiva.

Y en estos momentos de incertidumbre, estos momentos en que las empresas de un sector están golpeadas, es cuando merece la pena entender y estudiar dicho sector. Porque como decíamos al principio, a río revuelto, ganancia de pescadores.

Eso es precisamente lo que viene en los próximos posts. Tengo en el radar empresas como Visa, Mastercard, Adyen, Fiserv o PayPal — cada una operando en una capa distinta del ecosistema, con modelos de negocio distintos, y con amenazas distintas encima de la mesa. Ahora que tenemos el mapa general, podemos analizar cada empresa con mucho más contexto y más sentido.

Nos vemos en próximos posts. Suscribíos para recibirlos directamente en vuestro email.

Saludos!

English Version

Fishing in the Payments Ecosystem

Everything you need to know about the payments ecosystem before casting your line

They say troubled waters make for good fishing. And if you’ve been scrolling through social media or following economic and investment news lately, you’ve probably noticed that the payments ecosystem is pretty troubled right now:

• Fierce competition between traditional players and new entrants.

• Constant technological disruption.

• Cryptocurrencies and stablecoins pushing in from outside the system.

• Central banks making a mess with their own digital initiatives.

• Regulation and tax enforcement tightening.

• A2A payments (account to account), offered by some acquirers to pay directly from a bank account to another, bypassing much of the traditional ecosystem — leaving out Visa and Mastercard, for example.

• Closed-Loop payments, moving balances within the same app.

• Artificial intelligence redesigning processes that hadn’t been touched in decades.

• And even “tap to pay”, turning your phone into a payment terminal and cutting out hardware that used to be essential.

It’s a world that I’d say is simple at its core, because everything makes sense once you understand it. But right now it’s all a bit tangled. Something that in principle seems straightforward (”I give the order for money to move from source X, usually a bank account, and appear at destination Y, also usually a bank account”) has turned, after decades of development by different players, into a Frankenstein with many layers of functionality piled on top.

When you look at the whole ecosystem together, it feels like it could be massively simplified. And that, in my opinion, is the biggest structural threat to many companies in this sector. With the networks and technologies that existed two decades ago, everything made sense. But with the technology available today — incredibly powerful servers and a fully interconnected world with fast networks — much of the functionality looks redundant: layers that could be eliminated or collapsed, cutting out players that currently get paid just for sitting in the middle.

In upcoming posts I’m going to analyze several companies in this sector, and this article is a necessary introduction to understanding the payments ecosystem in general. In the following sections I’ll try to give a broad overview.

⚠️ Fair warning: this one’s a long read ⚠️

The next sections might be a bit dense, but trust me it’s worth the patience — once you have the context and understand the payments ecosystem, it’s much easier to understand the rationale behind each company: what each one does, how important it is, whether it has a real moat, and why its existence might be under threat.

B2B Movements: Bank-to-Bank

The foundation of this entire ecosystem is the transfer of capital between banks, from one bank account to another. All the other layers — the card networks of Visa or Mastercard, acquirers like Adyen or Fiserv, A2A payments, etc. — could ultimately be seen as add-ons designed to make these bank-to-bank transfers more convenient and secure.

“Transfer between bank accounts” sounds simple enough. In reality, there’s a massive infrastructure underneath that most people know nothing about.

The first thing to understand is that banks don’t communicate over the same internet you use to watch Netflix. The large banks (the so-called Tier 1: JPMorgan, Citigroup, BNP Paribas, Bank of America, Santander...) and central banks use what’s known as dark fiber: optical fiber strands leased long-term from telecommunications operators that are dedicated exclusively to that traffic. Nobody else passes through there. It’s literally a private tube of light from point A to point B. This isolates the network and largely prevents cybersecurity issues.

For international connections between the most important financial institutions, the story is similar but on a larger scale. SWIFT (the cooperative connecting more than 11,000 financial institutions worldwide) hasn’t laid its own cables across the Atlantic. However, it does operate its own private global network (SWIFTNet, a massive VPN based on MPLS technology), using the physical infrastructure of the major telecoms, but in a way that is completely segregated from the public internet: it uses the same physical channels (the same big internet cables) but has its own logical channels running “parallel” to the internet.

Some regions also have their own internal “highways.” In Europe, for example, there is SEPA (Single Euro Payments Area) and SEPA Instant (the Formula 1 version of SEPA). For transfers within Europe, SWIFT isn’t needed. In Spain specifically, there is Iberpay (SNCE).

Smaller banks that can’t afford this type of infrastructure do use the regular internet, but with military-grade encryption.

Correspondent Banking

Here comes an important concept — the one that really explains how the system works: correspondent banking. And something critical to understand about the ecosystem: SWIFT doesn’t move money. It moves messages. It’s essentially an ultra-secure messaging system between banks. The money itself moves through a different set of players: clearing houses.

• Interbank messaging.

• SWIFT.

• Clearing and settlement (the actual movement of money).

• In Europe: TARGET2 (operated by the Eurosystem, the central bank) and EBA Clearing.

• In the US: FedNow (the Federal Reserve’s relatively recent system) and The Clearing House (which operates CHIPS and RTP).

These settlement systems are known as RTGS (Real-Time Gross Settlement), and are operated mostly by the central banks themselves. They are the true plumbing of the financial system.

Large international banks hold accounts with each other. Santander has an account with real money deposited at JPMorgan (and vice versa). When Santander wants to send dollars to JPMorgan in New York, SWIFT is the channel through which the secure instruction travels: “Hey JPMorgan, this is Santander. Take $1,000 from my account with you and give it to the customer with account Y.” When Santander sends that instruction via SWIFT, JPMorgan simply makes a bookkeeping entry: it deducts the amount from Santander’s account and credits it to the final recipient. No physical money moves at that moment — just a balance adjustment.

And if the sending bank doesn’t have a direct relationship with the destination bank, it uses one or more intermediary banks that do. That’s what makes some international transfers slow (1–3 business days) and expensive: every intermediary in the chain needs to be paid.

Now, if Santander sends a lot of transfers to the US but few in the other direction, the account it holds at JPMorgan gradually empties. At some point it needs to be replenished with real money, and that happens through actual settlements between banks via the central bank systems: the RTGS systems already mentioned, like TARGET2 in Europe or Fedwire in the US. That’s where real central bank money moves — the “real” money at the bottom of the system.

The Limitations of the Bank-to-Bank System

The system we’ve just described works. And it’s been working for decades. But it has obvious friction: it’s expensive for international payments, doesn’t operate 24/7, and as soon as you cross your geographic zone, intermediaries, waiting times and fees start piling up. On top of that, this entire ecosystem is designed to move large amounts between institutions — not for you to pay for a coffee or buy something online.

And that’s precisely the origin of the next layer of the ecosystem. Visa and Mastercard weren’t born to compete with SWIFT or clearing houses. They were born to solve a completely different problem: how to let an individual pay at any merchant in the world, conveniently, quickly, and securely, without having to move money between bank accounts every time they buy something.

C2B Payments: From Consumer to Business

In a conventional card payment, the logical flow is:

Merchant → Gateway → Acquirer → Payment Networks (Visa/MC) → Issuing Bank

And the money flows in exactly the reverse direction during the settlement phase overnight, through the clearing houses we already covered.

Payment networks like Visa or Mastercard operate on private infrastructure — essentially a massive VPN with multi-million dollar contracts with telecom operators, completely segregated from the public internet. A delay of more than a few milliseconds beyond what’s agreed triggers significant penalties. What they offer, in essence, is speed in payment authorization and the infrastructure to make everything balance out afterwards.

The Magic of the Message in 200 Milliseconds

Networks like VisaNet or Banknet can process thousands of transactions per second with near-zero latency. When a customer swipes their card on a payment terminal, two phases take place operating at completely different speeds and in entirely different worlds:

Phase 1: Authorization (real time, under 1 second)

A message travels following the software flow mentioned above from the merchant to the issuing bank. The bank says “this customer has sufficient balance or credit, approved.” That OK blocks the funds in the customer’s account and assures the acquirer — and therefore the merchant — that the payment will go through. Not a single cent has moved yet, but the coffee is yours.

Phase 2: Clearing & Settlement (deferred, overnight)

This is where the “Card World” connects with the “Transfer World.”

Throughout the day, authorizations accumulate with each payment. At the end of the day, when all merchants close their terminals, acquirers go back to talk to Visa, Mastercard, or the relevant payment network. They package millions of transactions and send them to carry out the “netting”: the clearing between all banks. And through the RTGS systems of the central banks we already covered (TARGET2, FedWire), the actual movement of money takes place.

In other words: all payments are authorized and “pre-confirmed” throughout the day, and periodically a full accounting is done of all the money that actually needs to move. The payment networks perform the magic of that clearing.

“Right, Santander: your customers have spent 10M outside your bank. But merchants who bank with you have received 2M in purchases from other banks’ cards. Your net position for the day is negative 8 million. You need to pay up.”

And netting means exactly that: Santander doesn’t make thousands of transfers to pay the 10M and receive the 2M. It simply settles its net daily position — the 8 million — through the RTGS system it operates with its central bank.

What Are Acquirers Actually For?

After reading the previous sections, it might seem like acquirers are just a middleman sitting between the merchant and the payment network, skimming off the top. And yes, they are middlemen skimming off the top — but in the lines below we’ll see they genuinely earn their place.

You might wonder: if the payment network already does all this, why doesn’t the payment terminal give the customer’s data directly to Visa or Mastercard to authorize and later settle the payment? Why do Apple, Spotify or Uber pay millions to Adyen instead of connecting directly to Visa?

The main reason is that the world is chaotic. There are millions of merchants, each with their own particularities — their geographic region, their type of business, their customer base.

Connecting that chaos to the ultra-strict networks of Visa and Mastercard is a nightmare of engineering, regulation, fraud risk and treasury management. It’s like trying to plug a household toaster directly into a high-voltage power line. You need a transformer that adapts the power, manages the surges, makes sure the house doesn’t catch fire, and on top of that, groups everything into a single monthly bill. That transformer is the acquirer.

Translator

The acquirer acts as a translator: it takes Uber’s web format or the code from the corner shop’s terminal, cleans it up, encrypts it and hands it to the payment network perfectly packaged. If the payment network has a hiccup at any point, the acquirer holds the request and retries it, saving the merchant’s sale.

Smart Routing

If you connected directly to Visa, you’d be locked into their single route. A global acquirer like Adyen does Smart Routing: when you enter your card on an international website, the acquirer analyzes in milliseconds which bank issued the card, which country it’s from, which network is cheapest at that moment. If it sees that processing the payment through international Visa is going to cost 2% and has a 15% rejection probability due to fraud suspicion, it can decide to route that payment through a local network or a direct agreement with the issuing bank. This maximizes the conversion rate — meaning the payment doesn’t get declined by mistake — which is what really matters to an online merchant.

Fraud Shield

The acquirer takes on the credit and fraud risk of the merchant. That’s why you need a financial license to be an acquirer. Adyen assesses the financial health of Spotify or whichever airline before signing them up. If the airline goes bankrupt and thousands of customers claim refunds for their flights, Adyen has to pay that money out of its own pocket. Visa doesn’t want to take on the risk of dealing with millions of merchants; it prefers to hold a handful of regulated acquirers accountable and let them sort things out with each customer and merchant.

Treasury

Imagine a merchant sells a thousand coffees a day at €2 each. If Visa settled directly with the merchant, it would have to make a thousand micro-transfers daily — inefficient and extremely costly. The acquirer acts as a funnel: it receives a single block of net funds from the issuing banks covering millions of transactions, deducts its fee (plus Visa’s and the issuing bank’s), and pays the merchant one clean daily or weekly payment. The merchant reconciles nothing; they just see one daily deposit in their account.

U2U Payments: User-to-User (P2P Peer-to-Peer)

The third major block of the ecosystem is the one closest to the everyday user: sending money to a friend to split a dinner bill, divide rent, or settle a debt. In theory it’s the simplest use case of all — two people, two accounts, one amount. In practice, for decades it was also the most inconvenient, because it required knowing the other person’s IBAN, waiting a business day, and hoping you didn’t mistype a digit.

What has changed in recent years is that these payments no longer need to go through card networks. They use other banking rails we already covered directly — SEPA Instant, RTGS systems, Open Banking APIs — but with a software layer on top that makes the process as fast and simple as sending a WhatsApp message.

The most relevant examples globally:

Bizum (Spain): Runs on Iberpay and SEPA Instant. Instead of sharing your IBAN, you use your phone number. The money arrives in seconds.
Zelle (US): A consortium created by the major American banks to compete with third-party apps. Operates directly between bank accounts, with no intermediary balance.
Venmo / Cash App (US): A different model — the money sits in a balance inside the app itself (what’s called “closed-loop”) until the user decides to transfer it to their bank. More flexible for the user, but the money isn’t in your actual bank account until you move it out.
Pix (Brazil): Created directly by the Brazilian Central Bank. One of the most remarkable success stories globally — in just a few years it became the dominant payment method in Brazil, displacing both traditional transfers and cash.
UPI (India): India’s Unified Payments Interface is probably the largest P2P ecosystem in the world by transaction volume. A textbook case of how regulatory intervention can accelerate mass adoption of digital payments.

What’s interesting about this layer from an investor’s perspective is that it’s where the threat to the traditional ecosystem is most clearly visible. These systems don’t pay interchange fees to Visa or Mastercard. They don’t need acquirers. They go straight from account to account. And in the countries where they’ve truly taken off — Brazil, India, China with WeChat Pay and Alipay — the penetration of traditional card networks has been structurally limited.

Conclusion: The Map Before the Territory

If you’ve made it this far, congratulations — and thank you for the patience. It was a long read, I know. But it was a necessary one.

The payments ecosystem is one of those sectors where it’s very easy to read headlines (”Visa earns X billion,” “PayPal loses market share,” “stablecoins are going to kill the banks”) without really understanding what’s happening underneath. And without that context, it’s very hard to know whether the headline matters, whether the risk is real, or whether it’s just noise.

What we’ve seen in this article is that the ecosystem has three clearly differentiated layers. The B2B layer, which is the real plumbing of the financial system — SWIFT, correspondent banks, clearing houses — and which has worked for decades but carries obvious friction in international payments. The C2B layer, where Visa and Mastercard built their empire precisely to solve what the B2B system couldn’t: letting a consumer pay at any merchant in the world conveniently and securely. And the P2P layer, the newest one and the one most clearly threatening the layers above it, because it goes directly from account to account without needing card networks or acquirers.

And on top of all this: disruption on every front. A2A payments, stablecoins, Open Banking, central bank initiatives. All of them attacking one of these layers. All of them with real arguments. None of them, so far, with a decisive victory.

And in these moments of uncertainty — moments when the companies in a sector are under pressure — is exactly when it’s worth understanding and studying that sector. Because as we said at the beginning, troubled waters make for good fishing.

That’s precisely what’s coming in the next posts. On my radar: Visa, Mastercard, Adyen, Fiserv, and PayPal — each operating in a different layer of the ecosystem, with different business models, and with different threats on the table. Now that we have the general map, we can analyze each company with much more context and clarity.

See you in the next one. Subscribe to receive them directly in your inbox.

Cheers!

En el anterior post de Poetic Street ya repasé algunos argumentos de la narrativa actual contra Microsoft:

• CapEx desbocado por la inversión en centros de datos para la IA

• Dependencia de OpenAI

• SaaSmageddon

Y ya di mi opinión contra esta narrativa negativa, al menos en lo que se refiere a la dependencia de OpenAI y el SaaSmageddon. Si te interesó aquel artículo, no dudes en suscribirte para recibir más contenido similar.

Suscríbete ahora

En este post voy a ser más conciso y directo, e intentaré dar mi punto de vista sobre el problema del CapEx, que afecta no solo a Microsoft, sino a todos los hiperescaladores.

Valor Intrínseco de Microsoft

Antes de empezar, que quede claro, como siempre, que nada de esto es consejo de inversión, que puedo haber metido la pata hasta el fondo con los cálculos que voy a poner a continuación y, en resumen, que te imagines que soy un loco que pasa por la calle gritando cosas irracionales sobre bolsa y economía; que desconfíes de todo lo que leas y que hagas tu propio análisis o, al menos, contrastes el mío y me pongas la cara colorada en los comentarios si es necesario.

Dicho esto, vamos directos al ladrillazo y vamos a tomar algunos datos financieros anuales de Microsoft desde 2019 hasta 2025:

Como vemos en la captura, mi hoja de Google Sheets calcula el crecimiento medio anual del Revenue y el margen medio de conversión de Revenue a Free Cash Flow (FCF) en base a los datos históricos desde 2019 hasta 2025:

• Crecimiento medio anual del Revenue: 14,44%

• Margen medio de conversión Revenue a FCF: 30,28%

Si realizamos un descuento de flujos de caja con estos parámetros, un WACC del 8% y una tasa de crecimiento perpetuo del 3%, obtenemos un valor intrínseco de 382 dólares:

Está cerca del precio actual de la acción de Microsoft en el momento de escribir este post.

¿Y qué pasa con las toneladas de CapEx?

La estimación anterior suena muy bonita si Microsoft siguiera siendo una empresa Asset-Light que gasta poco y gana mucho.

Pero mantenerse en la carrera por la IA, con toda esa inversión en centros de datos y todas esas tarjetas gráficas friéndose cada vez que alguien le pregunta el horóscopo a ChatGPT por la mañana, cambia completamente el escenario.

Microsoft ya ha anunciado de forma oficial que estiman un gasto en CapEx de 190000 millones de dólares en 2026.

Aquí se abren varios escenarios futuros.

A) Escenario “Guay”: CapEx se estabiliza, ingresos se disparan

Satya Nadella defiende que el ritmo actual de CapEx se debe a que están construyendo los cimientos del nuevo stack tecnológico y que, a partir de 2027, a medida que la oferta de chips y componentes se estabilice, la presión sobre el CapEx cederá.

Además, las inversiones actuales en centros de datos no tardarán en generar ingresos. El consenso de analistas espera que para 2027 el CapEx empiece a estabilizarse en torno a los 130000 millones de dólares anuales, mientras que el flujo de caja operativo explotará al alza a medida que Azure (que crece al 40%) y las licencias de Copilot maduren y escalen en márgenes.

En este escenario, el FCF rebotaría con fuerza en 2027 y 2028.

B) Escenario “La Hemos Cagado”: CapEx incontrolable

Un edificio o una red eléctrica se deprecian en 20 o 30 años, pero los servidores de IA y los clusters de GPUs se quedan obsoletos en apenas 3 a 6 años debido al avance de los nuevos modelos.

Si en 2027 o 2028 Microsoft se ve obligada a sustituir de golpe los chips actuales por las siguientes generaciones para no perder cuota frente a Google o AWS, el CapEx no bajará.

Si el CapEx se mantiene estructuralmente por encima de los 150000 millones anuales y la demanda de servicios de IA no escala al mismo ritmo, los márgenes operativos (que ya estiman que bajarán al 44% el próximo trimestre) y el retorno sobre el capital invertido (ROIC) se resentirán de forma permanente.

Y aquí está la clave: no todo CapEx destruye valor. Si estas inversiones generan suficiente demanda futura y se monetizan adecuadamente, pueden resultar extraordinariamente rentables. El problema aparece cuando la inversión crece más rápido que la capacidad de generar retornos sobre ella.

C) Escenario “La Hemos Cagado Mucho”

En este escenario aprovecharíamos la capacidad de computación de los cerebros de un montón de humanos conectados a una matriz inmensa de procesamiento, reduciendo el gasto en CapEx enormemente... ¡Perdón! ¡Ya paro! Seamos serios, que nos jugamos la pasta.

Valor Intrínseco para Posibles Escenarios Futuros

Seguramente hay maneras mucho más avanzadas de estimar el valor intrínseco de Microsoft, incluyendo cifras exactas de estimaciones de CapEx futuro y utilizando métodos mucho más complejos.

Pero entender cómo está cambiando la empresa y poder realizar descuentos de flujo de caja con distintos parámetros para diferentes escenarios futuros probablemente ya nos da una ventaja sobre una buena parte de los inversores del mercado.

He utilizado mi hoja de cálculo jugando con diferentes valores de “Crecimiento medio anual del Revenue” y “Margen medio de conversión Revenue a FCF”.

Básicamente, jugando con el crecimiento medio anual del Revenue podemos estimar si la empresa incrementará o no sus ganancias gracias a la IA. Y con el margen medio de conversión Revenue a FCF podemos estimar si el CapEx se va de madre o no.

Además, conviene distinguir entre dos tipos de CapEx. Por un lado está el CapEx defensivo, destinado a mantener la competitividad frente a Google, Amazon o futuros competidores. Por otro lado está el CapEx ofensivo, cuyo objetivo es capturar una parte significativa del crecimiento futuro de la IA.

El primero puede verse como un coste necesario para no quedarse atrás. El segundo puede convertirse en una enorme fuente de creación de valor si la demanda termina justificando las inversiones actuales.

En la siguiente tabla se muestra el valor intrínseco para distintos escenarios:

La Bola de Cristal y Conclusión

Podemos leer cientos de artículos sobre IA, comparativas, pronósticos, expertos, etc. Pero la realidad es que hay momentos en los que la incertidumbre es inevitable.

Lo ideal sería encontrar buenas empresas, baratas y sin incertidumbre. Pero eso no suele pasar.

Si ahora encontramos Microsoft a los 385$ en los que cotiza en el momento de escribir este post, se debe precisamente a esta incertidumbre, a no saber qué parámetros utilizar para nuestros modelos y nuestros cálculos.

Mi humilde opinión, como inversor mediocre y como informático un poco menos mediocre, es que la IA es lo más “mágico” que se ha hecho, al menos en el campo de la informática, hasta el momento.

Internet fue brutal, pero no es tan difícil comprender su funcionamiento: no parece tan “por arte de magia”. La Blockchain es una genialidad, pero es relativamente fácil de entender. Los algoritmos para recrear mundos 3D en videojuegos, los virus informáticos y las herramientas para combatirlos...

Hay cientos de grandes invenciones en el mundo de la informática, pero jamás pensé que vería algo como la IA actual.

Creo que la carrera por la IA depende principalmente de dos cosas: capacidad de procesamiento y datos para entrenarla (y seguramente también regulación por parte de los gobiernos).

En mi opinión, va a ser muy difícil para nuevos jugadores competir en el campo de los centros de datos, y eso ofrece una gran barrera de entrada.

Pero además, pocas empresas tienen la cantidad de datos acumulados durante décadas como la tienen Google, Microsoft o Meta; y, sobre todo, la capacidad de ampliar esa cantidad de datos día tras día para entrenar la IA casi en tiempo real.

Y por otra parte, no sé vosotros, pero yo me declaro IA-dependiente absoluto, como en el pasado lo era de Google Search.

Y creo que llegará un momento en el que la IA sea tan necesaria para la humanidad como lo son el agua o la electricidad. Hasta el punto de que las empresas que se hagan un hueco en el sector llegarán a convertirse en una especie de utilities de los datos, y sus ingresos serán ultra predecibles y seguros.

Con todo esto sobre la mesa, en Poetic Street Capital tenemos una posición en Microsoft (exactamente el 5% de la cartera) a un precio medio de compra de 400$.

Y seguiremos ampliando si el precio sigue cayendo.

Porque creemos firmemente que el uso de la IA tendrá un crecimiento exponencial y que Microsoft será capaz, de una manera u otra, de generar retornos atractivos sobre las enormes inversiones que está realizando hoy.

Ya sea mediante mejoras en hardware, eficiencia energética, nuevos modelos de negocio o economías de escala, creemos que una parte significativa del CapEx actual terminará convirtiéndose en capacidad productiva altamente rentable.

Como siempre, nada de esto es consejo de inversión. Puedo estar equivocado de principio a fin, desde los cálculos realizados para obtener el valor intrínseco hasta todo el resto de la tesis.

Debéis hacer vuestro propio análisis y tomar vuestra propia decisión.

Si os ha gustado el artículo y os ha aportado valor, suscribíos para que os lleguen futuros artículos directamente a vuestro correo electrónico.

Y, por supuesto, cualquier comentario o corrección es bienvenido.

Se aceptan trolls y haters de todos los colores.

¡Saludos y un fuerte abrazo!

English Version

Microsoft and CapEx: The Hyperscaler Dilemma

Growing Without Dying of Success

In the previous Poetic Street post, I already went through some of the arguments behind the current bearish narrative surrounding Microsoft:

• Massive CapEx driven by AI data center investments

• Dependence on OpenAI

• SaaSmageddon

And I already shared my opinion on that negative narrative, at least regarding OpenAI dependency and the SaaSmageddon thesis.

In this post, I’ll be more concise and direct, and I’ll try to give my perspective on the CapEx issue, which affects not only Microsoft but all the hyperscalers.

Microsoft’s Intrinsic Value

Before we begin, let me make it clear—as always—that none of this is investment advice, that I may have completely messed up the calculations you’ll see below, and that, in short, you should imagine I’m a random lunatic shouting irrational things about stocks and economics in the middle of the street.

Take everything with a healthy dose of skepticism, do your own research, and feel free to make me look like an idiot in the comments if necessary.

With that out of the way, let’s jump straight into the financial nerd stuff and look at Microsoft’s annual financial data from 2019 through 2025:

As shown in the screenshot, my Google Sheets model calculates the average annual Revenue growth rate and the average Revenue-to-Free-Cash-Flow (FCF) conversion margin based on historical data from 2019 to 2025:

• Average annual Revenue growth: 14.44%

• Average Revenue-to-FCF conversion margin: 30.28%

If we run a discounted cash flow model using those parameters, an 8% WACC, and a 3% perpetual growth rate, we obtain an intrinsic value of $382 per share:

That’s fairly close to Microsoft’s current share price at the time of writing.

But What About the Mountains of CapEx?

The previous valuation sounds great if Microsoft remains an asset-light business that spends little and earns a lot.

But staying competitive in the AI race—with all those data centers and all those GPUs melting every morning because someone asks ChatGPT for their horoscope—changes the picture entirely.

Microsoft has already officially announced an estimated CapEx spend of $190 billion in 2026.

At this point, several future scenarios become possible.

A) The “Everything’s Fine” Scenario: CapEx Stabilizes, Revenue Explodes

Satya Nadella argues that the current pace of CapEx spending is driven by the need to build the foundations of the new technology stack and that, starting in 2027, as chip and component supply stabilizes, pressure on CapEx will begin to ease.

In addition, today’s investments in data centers should begin generating revenue relatively quickly. Analyst consensus expects CapEx to stabilize around $130 billion annually by 2027, while operating cash flow accelerates significantly as Azure (currently growing around 40%) and Copilot subscriptions mature and scale profitably.

Under this scenario, FCF would rebound strongly in 2027 and 2028.

B) The “We Screwed Up” Scenario: Uncontrolled CapEx

A building or an electrical grid may depreciate over 20 or 30 years, but AI servers and GPU clusters can become obsolete in as little as 3 to 6 years due to the rapid pace of technological advancement.

If Microsoft is forced in 2027 or 2028 to replace large portions of its current infrastructure with next-generation hardware just to remain competitive against Google or AWS, CapEx may never come down.

If CapEx remains structurally above $150 billion annually and demand for AI services fails to scale at the same pace, operating margins (which management already expects to decline to around 44% next quarter) and return on invested capital (ROIC) could suffer permanently.

And here’s the key point: not all CapEx destroys value. If these investments generate enough future demand and can be monetized effectively, they may turn out to be extraordinarily profitable. The problem arises when investment grows faster than the company’s ability to generate returns on that investment.

C) The “We Really Screwed Up” Scenario

In this scenario, we would leverage the computing power of millions of human brains connected to a giant processing matrix, dramatically reducing CapEx requirements...

Sorry. I’ll stop.

Let’s get serious. We’re talking about real money here.

Intrinsic Value Under Different Future Scenarios

There are undoubtedly far more sophisticated ways to estimate Microsoft’s intrinsic value, incorporating detailed future CapEx forecasts and more advanced valuation techniques.

But simply understanding how the business is evolving and modeling different future scenarios through discounted cash flow analysis probably already gives us an edge over a large portion of the market.

I’ve used my spreadsheet to test different assumptions for “Average Annual Revenue Growth” and “Revenue-to-FCF Conversion Margin.”

By adjusting Revenue growth, we can estimate whether AI will significantly increase Microsoft’s earnings power. By adjusting the Revenue-to-FCF conversion margin, we can estimate whether CapEx remains manageable or spirals out of control.

It’s also worth distinguishing between two different types of CapEx.

On one hand, there is defensive CapEx, aimed at maintaining competitiveness against Google, Amazon, and future challengers.

On the other hand, there is offensive CapEx, aimed at capturing a significant share of future AI growth.

The former may simply be the cost of staying in the game. The latter could become a tremendous source of value creation if demand ultimately justifies today’s investments.

The following table shows intrinsic value estimates under different scenarios:

Crystal Ball Time: Conclusion

We can read hundreds of articles about AI, comparisons, forecasts, expert opinions, and endless debates.

But the reality is that uncertainty is sometimes unavoidable.

Ideally, we’d find great businesses trading at cheap prices with no uncertainty attached.

That rarely happens.

If Microsoft is currently trading around $385 per share, it’s precisely because of this uncertainty. Investors simply don’t know which assumptions to use in their models.

My humble opinion, as a mediocre investor and a somewhat less mediocre software engineer, is that AI is the most “magical” thing humanity has built so far in the field of computing.

The Internet was revolutionary, but its underlying mechanics are relatively easy to understand. Blockchain is brilliant, but it’s also fairly straightforward conceptually. 3D rendering algorithms, malware, cybersecurity tools...

There have been countless incredible innovations in computing.

But I never thought I’d witness something quite like modern AI.

I believe the AI race depends primarily on two things: computing power and training data (plus, probably, government regulation).

In my opinion, it will be extremely difficult for new entrants to compete in the data center business, and that creates a significant barrier to entry.

Furthermore, very few companies possess the massive datasets accumulated over decades that Google, Microsoft, and Meta have built—and, more importantly, the ability to keep expanding those datasets every single day to train AI systems in near real time.

And on a personal level, I don’t know about you, but I’ve become completely AI-dependent, much like I once depended on Google Search.

I believe there will come a point when AI becomes as essential to humanity as electricity or running water.

The companies that successfully establish themselves in this space may eventually become a kind of data utility, with highly predictable and resilient revenue streams.

With all of that in mind, Poetic Street Capital currently holds a position in Microsoft (exactly 5% of the portfolio) at an average purchase price of $400 per share.

And we’ll continue adding if the stock keeps falling.

Because we firmly believe that AI adoption will grow exponentially and that Microsoft will ultimately generate attractive returns on the enormous investments it is making today.

Whether through hardware improvements, energy efficiency gains, new business models, or economies of scale, we believe a meaningful portion of today’s CapEx will eventually become highly productive and profitable capacity.

As always, none of this is investment advice.

I could be completely wrong—from the intrinsic value calculations all the way to the broader investment thesis.

Do your own research and make your own decisions.

If you enjoyed this article and found it useful, consider subscribing so future posts land directly in your inbox.

And of course, comments, corrections, trolls, and haters of all kinds are always welcome.

Cheers, and thanks for reading!

Hace unas semanas empecé a ver la serie televisiva “Perdidos” (sí, lo sé, no estoy a la última en series). He estado enganchadísimo. Y de buenas a primeras, no sé cómo ni por qué, al que podríamos considerar el protagonista principal (Jack), empezó a caerme realmente mal. ¿No os pasa esto con algún personaje de una serie de vez en cuando? O con algún youtuber, o sin ir más lejos con algún amigo o un compañero de trabajo. O quizás te está pasando ahora mismo conmigo. Esa persona que te cae bien, y de repente, sin venir a cuento, no le soportas. Un microgesto, quizás un pequeño comentario, posiblemente algo casi imperceptible, y piensas “menudo engreído”. Y ahora cada vez que le ves, se te tuerce el día.

No soy ningún gurú de la bolsa, pero llevo el tiempo suficiente en esto para saber que los mercados hoy en día se parecen bastante a lo que os cuento. Sin que haya una razón de peso, una empresa te empieza a caer mal. No sabes ni por qué.

Es más, seguramente los inversores más inexpertos lo entendemos incluso mejor que aquellos que llevan un porrón de años. Porque la dinámica del mercado de hoy en día posiblemente no tiene nada que ver con la de antes: no se mueve tanto por razonamientos coherentes y por fundamentales, sino por sensaciones y narrativas fugaces. Sentimientos corriendo a toda velocidad por las redes. Un mercado lleno de inversores que son auténticos “futboleros”, que sienten los colores. Compran Google, o Meta, o Bitcoin, y se ponen la camiseta de su equipo. Y lo defienden con la vida si hace falta.

Por eso hoy en día tan importante es para una empresa que tenga buenos fundamentales, como que despierte simpatía en el inversor.

Un ejemplo es SpaceX. Claro que es posible que tenga un futuro tremendamente prometedor, que conquiste el espacio, que nos lleve a todos a Marte, o que construya la Estrella de la Muerte y Elon Musk se convierta en un Jedi. Pero la realidad es que al momento de escribir este post, SpaceX tiene pérdidas, capex astronómico y un modelo de negocio en construcción. Y aun así supera ahora mismo (con sus 2,1 trillones de dólares) en capitalización a Meta, una empresa esta última que genera unas cifras de ingresos obscenas.

Y esto se debe muy en parte a que “cae bien”. Los inversores aman y son fanáticos de Elon Musk como si fuera el mismísimo Michael Jackson.

Y luego está el caso contrario, los que “caen mal”. Meta precisamente suele ser uno de los sospechosos habituales. Hace unos años Mark Zuckerberg también quiso tener su momento de “¡eh chicos, yo también soy cool!”. Zuckerberg no pretendía conquistar el espacio ni construir civilizaciones avanzadas de elonmuskianos detrás de algún agujero negro. Zuckerberg pretendía construir lo que llamaban el Metaverso. A Meta se le sale la pasta de los bolsillos, y pretendieron invertir en un proyecto que, por otra parte, sonaba bastante alineado con el resto de sus líneas de negocio y con sus redes sociales (y honestamente, un proyecto en el cual la IA hubiera encajado muy bien, pues todo ese Metaverso podría haber sido perfectamente una especie de “guardería” para entrenar IAs en un entorno virtual).

En ese entonces, Meta cayó desde 380 dólares a los que cotizaba en septiembre de 2021, hasta tocar los 88 en noviembre de 2022. Una caída del 77% para una empresa que tenía el dinero por castigo.

Hace unos meses sucedió algo similar con Google. Con la aparición de la IA y el ChatGPT de OpenAI, Google empezó a caer realmente mal a los inversores. Google se había dormido en los laureles, decían, y la IA le había pasado por encima. Sus búsquedas tradicionales ya no valían nada. La ingente cantidad de datos recopilados durante décadas, sus algoritmos para mapear internet, su cloud, sus “other bets”… todo eso ya no valía nada. Google ya no era cool.

El nuevo chico popular del instituto era Sam Altman con OpenAI. Para ese entonces Microsoft se llevó el gato al agua y se hizo con una participación del 27% en OpenAI justo cuando más molaba: una inversión de unos 135.000 millones de dólares. Y a cambio, OpenAI se comprometió a gastar unos 250.000 millones de dólares en servicios de la nube de Microsoft hasta el año 2032. Así que Microsoft se convirtió en el amigo del chico popular del instituto, llegando a cotizar a un PER cercano a 40x.

Pues bien, Google, desde abril de 2025, cuando tocó los 140 dólares, ha llegado a subir un 190% hasta casi tocar los 410 dólares por acción en mayo de 2026.

En menos de un año la narrativa ha cambiado por completo. Solo hay que darse una vuelta por X para verlo.

Ahora Google es la bomba. Tiene su propio modelo de IA, cantidades ingentes de datos para entrenarlo, capacidad para alimentarlo con nuevos datos recopilados de internet casi en tiempo real, ha integrado la IA perfectamente en las búsquedas tradicionales, fabrica sus propios chips, su cloud no para de crecer, Waymo (al menos de momento) es líder indiscutible en operaciones comerciales de robotaxis en EEUU (con más de 450.000 viajes pagados a la semana), Android, Chrome, participaciones en SpaceX y Anthropic... Y es cierto que Google es la bomba, de hecho es posiblemente la empresa que más me gusta de todo el índice. Pero la nueva narrativa ha puesto sus valoraciones por las nubes.

Mientras tanto, al contrario, Microsoft ha perdido popularidad. Se encuentra en su valoración más baja en una década porque la nueva narrativa es que la IA va a erosionar el modelo de suscripción por usuario sobre el que se construyó Microsoft (si hay menos trabajadores en las empresas, sustituidos por la IA, harán falta menos licencias de productos de Microsoft) y que Microsoft tiene demasiada dependencia de OpenAI. Por su parte, OpenAI ha dejado de ser el chico listo de la clase para dar paso a los flamantes modelos de Claude que están en boca de todos.

¿Tiene fundamento toda esta nueva narrativa en la que Google es guay, Claude va a convertirse en Skynet, y Microsoft va directo a la quiebra? Nadie puede saberlo, pero mirando los números y líneas de negocio de Microsoft, no parece que vaya a la quiebra.

Tanto Google como Microsoft han empezado a gastar toneladas de dólares en Capex para hacerse un hueco en la carrera de la IA. Sin embargo, el mercado lo ha celebrado o al menos se lo ha tolerado a Google, mientras que castiga duramente a Microsoft o Meta. Probablemente, si no estuvieran invirtiendo para competir en este campo, también habrían sido castigados, como pasó con Apple. O serían acusadas de que la IA está acabando con sus negocios principales, como le pasó al propio Google cuando se decía que la IA acabaría con las búsquedas tradicionales y que Google era la nueva Kodak.

¿Depende Microsoft de OpenAI?

Los acuerdos con OpenAI fueron el catalizador que dio a Microsoft una ventaja competitiva inicial, y aun a día de hoy, OpenAI sigue siendo la apuesta principal de Microsoft para competir en este campo.

No podemos engañarnos: en una carrera donde se compite sin piedad, como es la carrera actual en el cloud y la IA, que OpenAI fallara sería un buen palo para Microsoft. No solo por lo que supondría a nivel tecnológico, sino porque muchos lo interpretarían como la confirmación de la narrativa anti-Microsoft reinante.

Pero hay que entender que el verdadero negocio de Microsoft es la infraestructura y la distribución: cada vez que una empresa decide usar cualquier modelo de IA dentro del entorno empresarial, necesita la seguridad, los datos y la potencia de cálculo que pocas nubes proveen, y Azure es una de las principales.

A pesar de que OpenAI es la apuesta principal de Microsoft, han invertido agresivamente en el desarrollo interno de familias completas de modelos de lenguaje propios en dos frentes diferenciados: la serie Phi (como Phi-3, con modelos pequeños enfocados en una alta eficiencia, baja latencia y costos reducidos), y modelos fundacionales propios, la familia MAI (en la carta a los accionistas de finales de 2025, Satya Nadella anunció la introducción formal de sus primeros modelos fundacionales desarrollados y entrenados de extremo a extremo de manera 100% interna: MAI-1, MAI-Voice-1 y MAI-Image-1).

Por otra parte, Microsoft tiene la plataforma abierta y multimodelo Azure AI Foundry, un ecosistema que funciona como una tienda de IA para empresas con un catálogo de más de 11.000 modelos de inteligencia artificial de múltiples competidores y socios como Meta, Mistral, Cohere, DeepSeek, xAI, entre otros.

Por último, Microsoft está monetizando la IA en el software empresarial del día a día, integrando agentes y asistentes automatizados en toda su suite (Windows, Office 365, LinkedIn, Dynamics y GitHub).

Si OpenAI llegara a colapsar o a perder la carrera frente a rivales como Anthropic o Google, Microsoft impulsaría los modelos alternativos de código abierto o comerciales de terceros integrados en su Azure AI Foundry (los clientes corporativos de Microsoft podrían cambiar a otro modelo puntero sin salir de la nube de Azure), además de acelerar su propia línea interna de modelos MAI-1.

¿Damos ya por muerto a OpenAI?

Dicho todo lo anterior, ¿podemos dar ya por muerto a OpenAI? Hace unos meses OpenAI era el TOP de la IA. Luego apareció Gemini y se decía que Google había sabido plantar cara a OpenAI. Ahora resulta que Claude de Anthropic es el TOP de verdad.

En los últimos tiempos hemos visto titulares propios de una película de Steven Spielberg:

• La IA de Claude aprende a mentir y borra logs para ocultar sus huellas.

• La IA de Claude se escapa de un entorno controlado (una sandbox) creando el exploit para fugarse por sí misma, a la vez que informaba al desarrollador de que ya estaba fuera.

• Vulnerabilidades ocultas por más de 20 años en sistemas críticos que son descubiertas ahora por Claude.

• El gobierno de EEUU censurando la salida al público de ciertas versiones de Claude por considerarlas un peligro para la seguridad nacional.

Seguramente nunca sabremos qué parte de hype y marketing hay en todo esto.

Que Claude funciona como un reloj suizo no se le escapa a nadie. Pero honestamente, utilizo a diario Gemini y OpenAI junto a Claude, y cualquiera de ellos me resuelve la mayoría de mis problemas. Además, hay muchos otros factores a tener en cuenta más allá de qué modelo es el más “inteligente” o potente: el precio por token, el precio global del servicio que ofrece cada empresa con todas sus ventajas, la facilidad de integración en empresas…

Además, pensemos: con la bestial inversión en capex que están haciendo tanto Google como OpenAI, Microsoft… ¿de verdad nos sorprendería que mañana Google salga con un nuevo modelo que nos deje a todos con la boca abierta? ¿O que OpenAI haya conseguido la ansiada AGI (Artificial General Intelligence)?

No creo que nadie pueda decir con certeza que uno de estos modelos va a comerse a los otros, por mucho titular sensacionalista y mucho hype que se le dé. Además, el pastel de la IA es inmenso, y habrá un trozo para cada actor que consiga entrar en este mercado. Y no quitemos importancia a la regulación y a los gobiernos. ¿Va a permitir el gobierno de Estados Unidos que una única empresa se haga con todo el pastel? ¿Va a dejar caer a un ariete de la IA como OpenAI? Si no ha dejado caer a empresas menos críticas para el país como Intel, menos lo permitirá con OpenAI.

¿Y qué pasa con el SaaSMageddon?

El SaaSMageddon. La hipótesis de que la IA y los agentes autónomos destruirán el modelo de negocio tradicional de software bajo suscripción.

(No sé si OpenAI está muerto, pero tengo que reconocer que hace unos memes cojonudos).

La premisa es simple: si un agente de IA puede programar, automatizar flujos de trabajo complejos y crear herramientas a medida en cuestión de minutos, ¿para qué va a seguir pagando una empresa millones de dólares en licencias anuales de Salesforce, ServiceNow o SAP?

La cuestión es: ¿es Microsoft otra víctima más del SaaSMageddon, o más bien podría convertirse en un win-win para la compañía?

Por un lado, Microsoft, con Azure, tiene el blindaje perfecto: la infraestructura. Tiene una de las pocas autopistas de la computación para la IA. Si una startup o un equipo de desarrollo decide reemplazar un software de terceros creando su propia solución con agentes de IA, los modelos posiblemente se ejecutarán en nubes como Azure (sería raro y posiblemente ineficiente que la empresa instale y gestione localmente sus modelos de IA). Para Microsoft, el SaaSMageddon no destruye el gasto tecnológico, sino que lo transfiere: los dólares que las empresas antes destinaban a licencias de software ahora los destinan a capex y consumo de computación en la nube.

Por otra parte, el SaaSMageddon ofrece una tremenda oportunidad a Microsoft para reinventar su suite de productividad: Microsoft 365 + Copilot. El riesgo teórico existe: si la IA automatiza la creación de documentos, hojas de cálculo y correos, el valor de la interfaz tradicional de Office disminuye. Sin embargo, Microsoft cuenta con un foso defensivo casi impenetrable: gestiona el acceso seguro y unificado a todo el histórico de datos, correos, reuniones, documentos... de una corporación, todo ello bajo estrictas normativas de seguridad y cumplimiento. Ningún agente de IA externo puede operar eficientemente en una gran empresa si no tiene acceso a todo el contexto organizativo, mientras que el ecosistema de Microsoft ya está integrado hasta el tuétano en la empresa. Al añadir Copilot de forma nativa en ese ecosistema, Microsoft está metiendo con calzador su IA en la empresa: no está esperando a ser “disruptada” por la IA, está liderando de forma forzada la transición del SaaS tradicional al AaaS (Agent-as-a-Service).

Mientras que el SaaSMageddon es una amenaza real y existencial para las compañías de software de tamaño medio y nicho, para Microsoft representa una oportunidad histórica de consolidación de mercado.

En los próximos meses, será importante revisar los earnings reports, no tanto por ver el número bruto de licencias tradicionales vendidas, sino por ver el crecimiento de los ingresos por consumo de Azure AI y la tasa de adopción de Copilot en clientes corporativos. Suscribíos si queréis permanecer informados, porque aquí en Poetic Street Capital estaremos pendientes para hacer seguimiento de cualquier earnings report publicado. Si Microsoft logra transferir con éxito a sus usuarios del modelo de “pago por asiento” al modelo de “pago por valor añadido de IA”, el SaaSMageddon no hará más que ensanchar su foso competitivo frente al resto de la industria.

Conclusión

Al igual que hace unos meses la narrativa general era que OpenAI iba a pasar con su ChatGPT como un tractor por encima de las búsquedas tradicionales de Google. Al igual que hace unos años Mark Zuckerberg había perdido la cabeza dilapidando fortunas en ideas locas de metaversos y multiversos, e iba a llevar a Meta a la quiebra. Al igual que la narrativa más reciente es que Elon Musk va a crear una colonia de cyborgs en alguna estrella cercana y todo lo que pagues por SpaceX es poco.

Pues bien, al igual que todo lo anterior, la siniestra y tétrica nueva narrativa con Microsoft es que el SaaSMageddon y la decadencia de OpenAI van a terminar con una empresa que ha sabido lidiar con auténticas hecatombes y crashes en el pasado.

En próximos posts haremos un análisis financiero y un descuento de flujos de caja para Microsoft, con diferentes estimaciones de crecimiento de revenue y margen de conversión de revenue a free cash flow para diferentes posibles escenarios, de más a menos optimistas.

Actualmente en Poetic Street Capital ya dedicamos un 5% de nuestra cartera a Microsoft, y consideramos que cualquier precio por debajo de 400 dólares es una ganga para esta pedazo de empresa. Suscribíos para ser notificados de cualquier nuevo análisis o comentario sobre la empresa.

Como siempre, todo esto no es recomendación de compra o de venta. Es solamente nuestro punto de vista, y un seguimiento de lo que en Poetic Street Capital hacemos con nuestras inversiones. Todos vosotros sois responsables de hacer vuestro análisis y tomar vuestras decisiones. Y por supuesto, aquí estamos todos para aprender: si no estáis de acuerdo con nuestro punto de vista, o detectáis cualquier error, cualquier comentario es bienvenido, ¡incluso si viene de un trol!

¡Saludos!

English Version

MSFTgeddon: The SaaSMageddon and the Decline of OpenAI. Why I’m Loading Up on Microsoft Below $400.

Markets fall in and out of love with companies for no good reason. Microsoft is having its Meta moment — and the bull case is hiding in plain sight.

A few weeks ago I started watching Lost (yes, I know, I’m not exactly up to date on TV series). I got completely hooked. And then, out of nowhere, for no reason I could explain, I started to genuinely dislike what you could call the main protagonist — Jack. Does that ever happen to you with a character in a series? Or with a YouTuber, or for that matter with a friend or a coworker? Or maybe it’s happening to you right now with me. Someone you like just fine, and then suddenly, out of the blue, you can’t stand them. A micro-gesture, a small comment, something almost imperceptible, and you think “what an arrogant prick.” And now every time you see them, your day is ruined.

I’m no stock market guru, but I’ve been around long enough to know that today’s markets feel a lot like this. For no particular reason, a company just starts to rub you the wrong way. You can’t even say why.

And honestly, the less experienced investors probably understand this better than those who’ve been at it for decades. Because the market dynamics today may have nothing in common with the ones from before: it doesn’t move so much on coherent reasoning and fundamentals, but on fleeting feelings and narratives. Emotions racing across social media at full speed. A market full of investors who are pure football fans — they feel the colors. They buy Google, or Meta, or Bitcoin, and they put on the jersey. And they’ll defend it with their lives if they have to.

That’s why today it’s just as important for a company to have solid fundamentals as it is to be likeable.

Take SpaceX. Sure, it may have an incredibly promising future — conquer space, take us all to Mars, build a Death Star with Elon Musk as a Jedi. But the reality is that at the time of writing, SpaceX is losing money, has astronomical capex, and a business model still under construction. And yet it now (with $2.1 trillion) surpasses Meta in market cap, a company that generates obscene revenue figures.

A big part of that is simply because SpaceX is liked. Investors love and worship Elon Musk like he’s Michael Jackson himself.

And then there’s the flip side — the ones people just don’t like. Meta is usually the prime suspect. A few years ago Mark Zuckerberg also wanted his “hey guys, I’m cool too!” moment. Zuckerberg wasn’t trying to conquer space or build advanced Elonmuskian civilizations behind some black hole. Zuckerberg wanted to build what they called the Metaverse — a project that, by the way, seemed reasonably aligned with the rest of his business lines and social networks (and honestly, one where AI could have fit perfectly, since the whole Metaverse could have been a kind of virtual playground for training AIs).

Back then, Meta fell from $380 where it was trading in September 2021 all the way down to $88 in November 2022. A 77% drop for a company that was drowning in cash.

A few months ago something similar happened with Google. With the arrival of AI and OpenAI’s ChatGPT, Google started falling seriously out of favor with investors. Google had fallen asleep at the wheel, they said, and AI had steamrolled right over it. Its traditional search was worthless. The enormous amount of data collected over decades, its algorithms for mapping the internet, its cloud, its “other bets”… all worthless. Google wasn’t cool anymore.

The new popular kid at school was Sam Altman with OpenAI. Microsoft, meanwhile, pulled off a masterstroke and secured a 27% stake in OpenAI right when it was at peak cool: an investment of around $135 billion. In exchange, OpenAI committed to spending roughly $250 billion on Microsoft cloud services through 2032. So Microsoft became the best friend of the popular kid, eventually trading at a P/E ratio close to 40x.

Well — Google, from its low of $140 in April 2025, has since rallied 190% to nearly touch $410 per share in May 2026.

In less than a year, the narrative has completely flipped. Just scroll through X for five minutes.

Now Google is the bomb. It has its own AI model, massive amounts of data to train it on, the ability to feed it new data collected from the internet almost in real time, has seamlessly integrated AI into traditional search, manufactures its own chips, its cloud keeps growing, Waymo (at least for now) is the undisputed leader in commercial robotaxi operations in the US (with over 450,000 paid rides per week), Android, Chrome, stakes in SpaceX and Anthropic... And yes, Google really is the bomb — it’s probably the company I like most in the entire index. But the new narrative has sent its valuation into the stratosphere.

Meanwhile, Microsoft has fallen out of fashion. It’s sitting at its lowest valuation in a decade because the new narrative is that AI will erode the per-user subscription model that Microsoft was built on (if AI replaces workers at companies, fewer people means fewer Microsoft licenses), and that Microsoft is too dependent on OpenAI. For its part, OpenAI has lost its “smartest kid in class” crown to Anthropic’s Claude models, which are now on everyone’s lips.

Does any of this new narrative hold up — Google is great, Claude is going to become Skynet, and Microsoft is headed for bankruptcy? Nobody can say for certain, but looking at Microsoft’s numbers and business lines, it doesn’t look like it’s going bankrupt anytime soon.

Both Google and Microsoft have started pouring enormous amounts of capex into the AI race. Yet the market has celebrated (or at least tolerated) Google’s spending, while punishing Microsoft and Meta hard for doing exactly the same thing. Ironically, if they weren’t investing to compete in this space, they’d probably be punished too — just like Apple was. Or they’d be accused of AI destroying their core businesses, like Google itself was when everyone said AI would kill traditional search and that Google was the new Kodak.

Is Microsoft dependent on OpenAI?

The OpenAI deals were the catalyst that gave Microsoft an initial competitive edge, and even today, OpenAI remains Microsoft’s primary bet in the AI race.

Let’s not kid ourselves: in a cutthroat competition like the current cloud and AI race, an OpenAI failure would be a real blow to Microsoft. Not just technologically, but because many would interpret it as confirmation of the anti-Microsoft narrative that’s already out there.

But here’s what matters: Microsoft’s real business is infrastructure and distribution. Every time a company decides to use any AI model in a corporate environment, it needs the security, data handling, and computing power that only a handful of clouds can provide — and Azure is one of the main ones.

Even with OpenAI as its primary bet, Microsoft has been aggressively investing in its own internal model families across two distinct tracks: the Phi series (like Phi-3, with small models focused on high efficiency, low latency and reduced costs), and its own foundational models — the MAI family (in its late 2025 shareholder letter, Satya Nadella announced the formal introduction of MAI-1, MAI-Voice-1 and MAI-Image-1, the first fully in-house, end-to-end trained foundational models).

On top of that, Microsoft has Azure AI Foundry — an open, multi-model platform that functions as an enterprise AI marketplace with a catalog of more than 11,000 AI models from multiple competitors and partners including Meta, Mistral, Cohere, DeepSeek, and xAI.

And finally, Microsoft is monetizing AI across its everyday enterprise software stack, embedding agents and automated assistants throughout its entire suite: Windows, Office 365, LinkedIn, Dynamics, and GitHub.

If OpenAI were to collapse or lose the race to rivals like Anthropic or Google, Microsoft would simply accelerate the open-source or third-party commercial alternatives already integrated into Azure AI Foundry — corporate clients could switch to a different leading model without ever leaving the Azure cloud — while simultaneously pushing its own MAI-1 internal model line harder.

Should we write off OpenAI already?

Given all of the above — can we actually write off OpenAI? A few months ago OpenAI was the TOP of AI. Then Gemini appeared and everyone said Google had successfully challenged OpenAI. Now apparently Anthropic’s Claude is the real TOP.

Recently we’ve seen headlines straight out of a Steven Spielberg movie:

• Claude’s AI learns to lie and deletes logs to cover its tracks.

• Claude’s AI escapes a controlled environment (a sandbox) by creating its own exploit to break out, while simultaneously informing the developer that it was already outside.

• Hidden vulnerabilities sitting undetected for over 20 years in critical systems now being discovered by Claude.

• The US government censoring certain Claude versions from public release on national security grounds.

We’ll probably never know how much of this is hype and marketing.

Nobody disputes that Claude works like a Swiss watch. But honestly — I use Gemini and OpenAI alongside Claude every single day, and any one of them solves the vast majority of my problems. And there are many other factors beyond which model is the most “intelligent” or powerful: price per token, the overall value of each company’s full service offering, ease of enterprise integration...

Besides, think about it: with the staggering capex that Google, OpenAI, Microsoft and others are pouring in — would any of us really be shocked if Google dropped a new model tomorrow that leaves everyone speechless? Or if OpenAI announced they’d achieved AGI?

Nobody can say with certainty that any one of these models is going to eat the others alive, regardless of how many sensationalist headlines and hype cycles we go through. The AI pie is enormous, and there’ll be a slice for every player that manages to get into this market. And let’s not underestimate regulation and governments. Is the United States government going to allow a single company to take the whole pie? Is it going to let an AI powerhouse like OpenAI go under? If it hasn’t let less strategically critical companies like Intel fail, it’s certainly not going to let OpenAI.

What about the SaaSMageddon?

The SaaSMageddon. The hypothesis that AI and autonomous agents will destroy the traditional subscription software business model.

(I don’t know if OpenAI is dead, but I have to admit they make some absolutely brilliant memes.)

The premise is simple: if an AI agent can code, automate complex workflows, and build custom tools in minutes — why would any company keep paying millions in annual licenses for Salesforce, ServiceNow, or SAP?

The question is: is Microsoft just another SaaSMageddon victim, or could this actually turn into a win-win for the company?

On one hand, Microsoft’s Azure gives it the perfect shield: the infrastructure. It owns one of the very few AI computing highways. If a startup or a development team decides to replace third-party software by building their own AI agent-powered solution, those models will almost certainly run on clouds like Azure (it would be unusual — and probably inefficient — for a company to install and manage AI models locally). For Microsoft, the SaaSMageddon doesn’t destroy tech spending, it redirects it: the dollars companies used to spend on software licenses now flow into capex and cloud computing consumption.

On the other hand, the SaaSMageddon presents a massive opportunity for Microsoft to reinvent its productivity suite: Microsoft 365 + Copilot. The theoretical risk exists: if AI automates the creation of documents, spreadsheets and emails, the value of the traditional Office interface diminishes. But Microsoft has a nearly impenetrable moat: it manages secure, unified access to a corporation’s entire history of data, emails, meetings, documents — all under strict security and compliance frameworks. No external AI agent can operate efficiently inside a large enterprise without access to the full organizational context, and the Microsoft ecosystem is already embedded into every corporate artery. By integrating Copilot natively into that ecosystem, Microsoft isn’t waiting to be disrupted by AI — it’s forcibly leading the transition from traditional SaaS to AaaS (Agent-as-a-Service).

While the SaaSMageddon is a real and existential threat for mid-size and niche software companies, for Microsoft it represents a historic market consolidation opportunity.

In the coming months, it will be important to watch earnings reports — not so much for raw traditional license numbers, but for Azure AI consumption revenue growth and Copilot adoption rates among enterprise clients. Subscribe if you want to stay informed, because here at Poetic Street Capital we’ll be tracking every Earnings Report as it drops. If Microsoft successfully transitions its users from a “pay per seat” to a “pay per AI value added” model, the SaaSMageddon will only widen its competitive moat against the rest of the industry.

Conclusion

Just like a few months ago the prevailing narrative was that OpenAI’s ChatGPT was going to steamroll Google’s traditional search like a freight train. Just like a few years ago Mark Zuckerberg had lost his mind burning fortunes on Metaverse and Multiverse pipe dreams and was going to drive Meta into the ground. Just like the more recent narrative that Elon Musk is going to build a colony of cyborgs near some neighboring star and no price you pay for SpaceX is too high.

Well — just like all of the above, the dark and gloomy new narrative around Microsoft is that the SaaSMageddon and the decline of OpenAI are going to bring down a company that has survived genuine catastrophes and market crashes in the past.

In upcoming posts, we’ll do a full financial analysis and discounted cash flow model for Microsoft, with different revenue growth and FCF conversion margin assumptions across a range of scenarios from most to least optimistic.

At Poetic Street Capital we currently allocate 5% of our portfolio to Microsoft, and we consider anything below $400 a bargain for this beast of a company. Subscribe to be notified of any new analysis or commentary on the stock.

As always, none of this is a buy or sell recommendation. It’s simply our perspective, and a running account of what we do with our own investments at Poetic Street Capital. You are all responsible for doing your own analysis and making your own decisions. And of course, we’re all here to learn — if you disagree with our take or spot any errors, all comments are welcome. Even from trolls!

Cheers!

Muchos inversores ignoran por completo los bonos y ni siquiera se molestan en entender su funcionamiento. Pero los bonos son posiblemente el activo más importante del mercado, el que más lo mueve y, en buena medida, el que condiciona toda la economía.

Predecir el comportamiento de este activo es posiblemente una de las tareas más difíciles para un inversor, junto con el comportamiento de las divisas (de hecho, invertir en bonos implica, de forma indirecta, invertir en divisas: de nada sirve comprar bonos de un país al 4% de interés si la divisa de ese país cae un 8%). Predecir el comportamiento de los bonos implica predecir factores macroeconómicos, políticos del país del bono, así como factores geopolíticos e internacionales. En mi humilde opinión, casi nadie puede ver con certeza hacia dónde irán los tipos de interés de un país.

Sin embargo, una cartera diversificada debe tener bonos que te protejan en ciertos escenarios y ciclos económicos, por ejemplo en una etapa prolongada de recesión o para amortiguar un crash por un cisne negro.

En mi cartera, considerando la situación actual donde la renta variable podría estar bastante sobrecalentada, tengo una posición considerable en activos fuera de la renta variable (oro, bonos, materias primas, liquidez...). En concreto, mi posición actual en bonos asciende a un sustancial 30% de la cartera.

Puesto que considero tremendamente difícil adivinar hacia dónde irán los tipos de interés y las divisas de cada país, toda mi posición en bonos la tengo a través de un ETF que sigue el Bloomberg Aggregate Bond Index. El Bloomberg Aggregate Bond Index está compuesto principalmente por bonos gubernamentales, corporativos y respaldados por hipotecas de alta calidad crediticia, los cuales se diversifican de forma equilibrada a través de diferentes sectores económicos, regiones geográficas y plazos de vencimiento (duración). Esta variedad de plazos y emisores permite capturar el comportamiento general de la renta fija segura, reduciendo el riesgo y ofreciendo una imagen clara de cómo se mueve el mercado de deuda a nivel global o regional.

iShares Core Global Aggregate Bond UCITS ETF

El ETF que utilizo para invertir en bonos es el ETF AGAC de BlackRock: iShares Core Global Aggregate Bond UCITS ETF:

https://www.blackrock.com/es/profesionales/productos/337224/ishares-core-global-aggregate-bond-ucits-etf

El ETF está diversificado en bonos gubernamentales y corporativos de distintos países y duraciones, predominando los bonos estadounidenses, y teniendo una duración media de 6–7 años.

Y el “Lambo” para cuándo

Efectivamente, los bonos no te van a hacer trillonario. Los bonos son un tipo de activo que nos ofrece estabilidad, seguridad y menor volatilidad en la cartera, obviamente con una rentabilidad mucho más ajustada.

El problema es que muchas veces, cuando compramos un ETF de este tipo, es muy difícil calcular qué rentabilidad podemos esperar a medio plazo, al menos para un inversor retail como muchos de nosotros. No porque sea difícil estimar la rentabilidad de un bono. Todo lo contrario: cuando compras un bono, sabes de antemano, siempre y cuando no se produzca un impago, la rentabilidad que puedes obtener (a diferencia de la bolsa, donde necesitas hacer una estimación). Es difícil porque la cantidad de activos (bonos) que componen el ETF es altísima.

En la propia web de BlackRock podemos descargar un documento CSV con todos los holdings que componen el ETF AGAC y podemos observar que son en torno a 20.000 bonos diferentes.

Si quisiéramos calcular qué rentabilidad exacta nos ofrecen los bonos que componen el ETF necesitaríamos obtener la rentabilidad exacta de cada uno de ellos hoy mismo (al precio de mercado actual de ese bono, no nos basta con saber su cupón), para luego calcular una media de todas las rentabilidades (una media ponderada según el peso de cada bono en el total del ETF).

Quizás grandes empresas como BlackRock pueden realizar este cálculo en tiempo real sin problemas, pero para inversores retail como la mayoría de nosotros, esto es una tarea titánica.

Cálculo de la rentabilidad aproximada

Puesto que calcular la rentabilidad exacta, como comentábamos en el apartado anterior, se escapa a nuestras capacidades, necesitamos idear un plan para hacer un cálculo aproximado.

Como vimos en las capturas un poco más arriba, BlackRock nos ofrece en su web el porcentaje de bonos de cada país, así como el porcentaje de bonos por duración.

Aquí obviamente hay bonos gubernamentales y corporativos, y las rentabilidades de cada uno son diferentes.

En nuestra aproximación lo que haremos será asumir que son bonos gubernamentales. Y lo que haremos será cruzar la tabla de peso por país con la tabla de peso por duración, para construir una matriz de pesos de cada par país-duración. Una vez construida dicha tabla, necesitaremos obtener la rentabilidad actual de los bonos para cada país y cada duración. Podríamos obtenerla, por ejemplo, en TradingView (si queremos ver la rentabilidad de los bonos de 30 años estadounidenses, buscaríamos el ticker US30Y. Para el de China a 20 años, CN20Y. Y así sucesivamente).

Puesto que tenemos 96 pares país-duración (habiendo tenido en cuenta solo los países que más pesan en el ETF), es un trabajo obtener manualmente cada rentabilidad (que además cambia con el precio del bono en el mercado). Así que hemos pedido a Gemini, la inteligencia de Google, que nos complete la tabla con las rentabilidades de cada país-duración.

En la captura anterior podemos observar que cada celda de la tabla corresponde al peso en el ETF de ese par país-duración y, entre paréntesis, la rentabilidad actual de esos bonos. Por ejemplo, el 1,16% del ETF corresponde a bonos estadounidenses de 0 a 1 años de duración, cuya rentabilidad ronda el 3,78%.

Con esta tabla, solo necesitamos calcular la media ponderada (en base a los pesos de cada par bono-duración en el ETF) de todas las rentabilidades.

Hemos pedido a Gemini que realice este cálculo y, según este, la rentabilidad media que nos ofrecen toda esta lista de bonos es aproximadamente del 3,61%.

Debemos tener en cuenta que hemos asumido que todos los bonos son gubernamentales y que, para cada par país-duración, hemos usado la rentabilidad del bono de ese país para esa duración. Los bonos corporativos ofrecen un extra de rentabilidad (porque también tienen un extra de riesgo frente a los gubernamentales). Así que posiblemente la rentabilidad real del ETF gire en torno al 4%.

Yield to Worst

A pesar de que hemos realizado este ejercicio para calcular una rentabilidad aproximada de nuestro ETF y para entender mejor en qué estamos invirtiendo, normalmente las empresas que emiten dicho ETF ya nos ofrecen algunas métricas que nos ahorran tener que hacer cálculos manualmente.

Si vamos al fact sheet del ETF AGAC, actualizado a abril de 2026, podemos ver algunas métricas que coinciden con nuestros cálculos:

• Average Weighted Maturity: 8.10 yrs

• Effective Duration: 6.15 yrs

• Yield to Worst: 3.83%

• Number of Holdings: 19976

El Yield to Worst (YTW) (que se podría traducir como “rentabilidad en el peor de los escenarios”) es una métrica clave en el mundo de la renta fija. Básicamente indica la mínima rentabilidad anualizada estimada que puedes esperar recibir de un bono o de una cartera de bonos (como este ETF), asumiendo que el emisor no quiebre.

Conclusión

Los bonos son un activo relativamente aburrido, y deben serlo, pero espero que este artículo no os haya resultado un ladrillo y que os haya ayudado a echar un poco de luz a un activo que muchos inversores desconocen o ignoran, pero que, en mi opinión, es importante no perder de vista y tener en nuestras carteras.

En cualquier caso, como siempre, esto no es un consejo de inversión, y cada uno debe estudiar y diseñar su propia cartera y decidir qué activos son apropiados para su perfil.

Si os ha gustado este artículo, suscribiros para recibir futuros posts en vuestro email. Saludos!

English Version.

Making Peace with Bonds: Your Portfolio’s Unsung Shock Absorber

Understanding bonds can stabilize your portfolio and save you from a few unpleasant surprises. Let’s go!

Many investors ignore bonds entirely and don’t even bother understanding how they work. Yet bonds are arguably the most important asset in the market — the one that moves markets and the broader economy more than anything else.

Predicting bond behavior is arguably one of the hardest tasks for any investor, right alongside predicting currency movements (and investing in bonds is implicitly investing in currencies: buying bonds from a country yielding 4% means nothing if that country’s currency drops 8%). Predicting bonds requires forecasting macroeconomic factors, country-specific political dynamics, and geopolitical conditions. In my humble opinion, almost no one can say with certainty where a country’s interest rates are headed.

That said, a well-diversified portfolio should include bonds that protect you in specific scenarios and economic cycles — think an extended recession, or a cushion against a black-swan crash.

In my own portfolio, given the current environment where equities may be quite overheated, I hold a substantial position in non-equity assets (gold, bonds, commodities, cash…). My current bond position stands at a meaningful 30% of the portfolio.

Since I consider it extremely difficult to predict where interest rates and currencies are heading, my entire bond exposure is through a single ETF that tracks the Bloomberg Aggregate Bond Index. This index is composed primarily of high-quality government, corporate, and mortgage-backed bonds, diversified across economic sectors, geographies, and maturities. That range of durations and issuers captures the overall behavior of safe fixed income, reduces risk, and provides a clear picture of how global (or regional) debt markets move.

iShares Core Global Aggregate Bond UCITS ETF

The ETF I use for bond exposure is BlackRock’s AGAC ETF: the iShares Core Global Aggregate Bond UCITS ETF:

https://www.blackrock.com/es/profesionales/productos/337224/ishares-core-global-aggregate-bond-ucits-etf

The ETF is diversified across government and corporate bonds from multiple countries and maturities, with a heavy tilt toward U.S. bonds and an average duration of around 6–7 years.

So, when Lambo?

Indeed, bonds aren’t going to make you rich overnight. Bonds offer stability, security, and lower volatility in a portfolio — naturally, at the cost of lower returns.

The challenge is that when you buy this type of ETF, estimating your expected medium-term return is genuinely tricky, at least for retail investors like most of us. Not because estimating bond returns is inherently difficult — quite the opposite. When you buy a bond, you know upfront (assuming no default) exactly what return you’ll get. That’s very different from equities, where you’re always estimating. The difficulty here is that the ETF holds an enormous number of bonds.

On BlackRock’s own website you can download a CSV file with all the ETF’s holdings — and you’ll find roughly 20,000 different bonds.

To calculate the exact yield of all the bonds in the ETF, you’d need the current yield of each one (at today’s market price, not just the coupon), then compute a weighted average based on each bond’s weight in the portfolio. Large institutions like BlackRock can do this in real time. For retail investors, it’s a Herculean task.

Estimating the Approximate Yield

Since calculating the exact yield is beyond reach for most of us, we need a practical approximation.

As shown in the screenshots above, BlackRock provides both the country weight breakdown and the duration weight breakdown on their website.

The portfolio includes both government and corporate bonds, each with different yields.

For our approximation, we’ll assume all bonds are government bonds. The method: cross the country weight table with the duration weight table to build a country × duration weight matrix. Once that’s done, we need the current yield for each country-duration pair — which you can look up on TradingView (e.g., US30Y for the 30-year U.S. Treasury, CN20Y for China’s 20-year bond, and so on).

Since there are 96 country-duration pairs (considering only the ETF’s most significant country weights), manually retrieving each yield is tedious — especially since yields shift with bond prices daily. So I asked Gemini, Google’s AI, to populate the table with the relevant yields for each country-duration pair.

In the table above, each cell shows the ETF weight for that country-duration pair, with the current yield in parentheses. For example, 1.16% of the ETF consists of U.S. bonds with a 0–1 year duration, currently yielding around 3.78%.

With this table, the only remaining step is a weighted average of all yields (weighted by each bond’s share in the ETF).

I asked Gemini to run this calculation. The result: the weighted average yield across all these bonds is approximately 3.61%.

Keep in mind: we assumed all bonds are government bonds and used the sovereign yield for each country-duration pair. Corporate bonds carry an additional yield premium (reflecting their extra risk). So the ETF’s actual real-world yield likely hovers around 4%.

Yield to Worst

Even though we did this exercise to estimate an approximate yield and better understand what we’re actually invested in, ETF providers typically publish metrics that save us from having to do the math manually.

If you check the fact sheet for the AGAC ETF, updated to April 2026, you’ll find metrics that align closely with our estimates:

• Average Weighted Maturity: 8.10 yrs

• Effective Duration: 6.15 yrs

• Yield to Worst: 3.83%

• Number of Holdings: 19,976

Yield to Worst (YTW) — roughly “return in the worst-case scenario” — is a key fixed income metric. It represents the minimum annualized return you can expect from a bond or bond portfolio (like this ETF), assuming no default by the issuer.

Conclusion

I hope this post has helped shed some light on an asset class that many investors overlook — yet one that, in my view, deserves a place in any thoughtful portfolio.

As always, this is not investment advice. Everyone should study and design their own portfolio based on their individual risk profile.

Thanks for reading, and if you enjoyed this post, subscribe below!

Some of the sector’s most established names have suffered significant multiple compression despite continuing to grow revenue and generate cash flow. The Global X Cybersecurity ETF (BUG) fell 17.6% in Q1 2026 alone. The market narrative is straightforward:

Generative AI will automate security work, reducing the need for specialized cybersecurity vendors.

It sounds logical.

It is also likely wrong.

A recent security incident at Meta may have exposed one of the biggest misconceptions currently embedded in the market’s AI narrative.

The lesson isn’t that AI systems are insecure.

The lesson is that every layer of AI automation creates entirely new categories of security risk that did not previously exist.

And somebody will have to secure them.

The Meta Incident Wasn’t Really About Instagram

Last week, reports emerged that attackers successfully manipulated Meta’s AI-powered support assistant into helping them take control of high-profile Instagram accounts.

The method was alarmingly simple. Attackers opened a conversation with Meta’s AI Support Assistant and asked it to link a target account to a new email address under their control. The chatbot sent a verification code to the attacker’s email; the attacker shared the code back with the bot, the chatbot displayed a “Reset Password” button, and the account was compromised.

Victims included the Instagram account associated with the Obama White House—briefly used to broadcast Iranian propaganda—the account of the U.S. Space Force’s Chief Master Sergeant, Sephora, and security researcher Jane Wong, among others. Meta patched the issue, though according to TechCrunch, attacks continued for days even after the company declared the problem resolved.

Back in December 2024, Meta had announced this assistant promising it would make account recovery “faster and simpler” and that it would be able to “reset your password securely.” Solutions, not just suggestions, the product page read.

Too efficient, as it turned out.

Most people looked at this and saw a chatbot failure.

I see something much larger.

The attackers didn’t discover a sophisticated memory corruption bug. They didn’t develop nation-state malware. They didn’t break modern cryptography.

Instead, they exploited a design mistake that is becoming increasingly common across corporate AI deployments:

An AI system was granted authority over a sensitive workflow.

The chatbot wasn’t merely answering questions. It was allowed to perform actions.

That distinction matters enormously.

The New Attack Surface: AI With Permissions

For decades, security teams protected applications.

Today they increasingly need to protect autonomous decision-making systems.

Traditional software behaves deterministically. An API endpoint follows predefined logic. A database follows predefined rules. A firewall enforces predefined policies.

Large language models are fundamentally different.

They operate probabilistically. They interpret instructions. They reason over context. And most importantly: they can be manipulated through language.

The moment an organization connects an AI agent to production systems, it creates a new attack surface unlike anything security teams have dealt with before.

The AI agent becomes a privileged intermediary between humans and infrastructure.

Attackers no longer need to exploit code.

Sometimes they only need to exploit judgment.

The Meta attack required no advanced technical knowledge. It required knowing how to talk to a chatbot. That has profound implications for who can launch these attacks and how often.

The Great Misunderstanding: Automation Does Not Eliminate Risk

Corporate executives are currently engaged in what we might call “Automation at Any Cost.”

Every workflow is being evaluated through a single lens: Can AI automate this?

Customer support. Software development. HR. Finance. Procurement. Legal review. Account recovery.

The assumption is simple: More automation = lower costs. That part is true.

What gets ignored is the second-order effect: More automation = more systemic risk.

Every AI agent introduced into a workflow creates:

• New identities

• New permissions

• New integrations

• New API dependencies

• New data exposure paths

• New attack vectors

• New failure modes

The AI layer doesn’t replace complexity. It adds another layer of complexity on top of existing systems.

Think of it as digital scaffolding. Companies are rapidly constructing AI-driven structures above their existing technology stacks. The scaffolding creates productivity. But it also creates new points of failure.

Meta is not a struggling legacy enterprise. It is one of the most sophisticated technology organizations on Earth, investing tens of billions in AI. If a company like that can accidentally create an exploitable recovery workflow, what does that imply for the average bank, insurer, retailer, hospital, or government agency currently rushing to deploy AI agents?

The answer is uncomfortable: Most organizations have barely begun thinking about these risks.

The Numbers Back the Thesis

The global cybersecurity market currently stands at roughly $219 billion in 2025 and is projected to reach nearly $700 billion by 2034, at a 13.8% compound annual growth rate.

But that is the baseline. The future could be significantly larger.

By 2029, agentic AI is expected to drive 15% of global cybersecurity budgets—nearly three times higher than current levels.

This projection reflects a growing recognition: threat actors are gaining the same AI capabilities as defenders, raising the speed and quality of attacks. The environment is getting harder, not easier, and spending will follow.

Wedbush analyst Dan Ives, one of the most followed voices in tech equities, has argued that “AI will be a major tailwind to the cybersecurity sector over the coming years as protection of use cases, data, and endpoints expands markedly.”

The market currently appears to be betting exactly the opposite.

The Coming Security Spending Wave

This is where the market may be making a category error.

Many investors assume AI will absorb functionality currently delivered by cybersecurity vendors. The opposite outcome appears more plausible.

Every dollar spent on AI automation may require additional spending on:

• Identity and access management

• Privileged access controls

• API security

• AI observability

• Runtime monitoring

• Data lineage tracking

• Prompt injection protection

• Model governance

• Agent authorization frameworks

• Human approval systems

AI adoption does not reduce security requirements. It expands them.

The more authority organizations give to AI systems, the more valuable security controls become.

The Cybersecurity Industry Is About to Gain a New Customer

Historically, cybersecurity vendors sold protection for: users, devices, networks, applications, and data.

Now a sixth category is emerging: Agents.

AI agents need identities. They need permissions. They need monitoring. They need behavioral controls. They need audit trails. They need containment mechanisms.

This is not a temporary trend. It is a new layer of enterprise infrastructure.

And every new infrastructure layer eventually develops its own security stack.

The cloud did. Mobile did. SaaS did. AI will too.

The Investment Implication

The current market narrative treats AI as a threat to cybersecurity spending. That framing may prove backwards.

AI is not removing the need for security. It is creating an entirely new security layer that didn’t previously exist.

That said, not all companies in the sector will benefit equally. The market is differentiating, and rightly so.

Several well-established names are worth examining:

CrowdStrike (CRWD) and Palo Alto Networks (PANW) carry the lowest execution risk given their scale. CrowdStrike recently pushed toward all-time highs; PANW’s “platformization” approach is widely seen as a game-changer for securing AI data pipelines. Both were up roughly 16% year-to-date through May while the broader software index (IGV) fell 16%—a significant divergence that speaks to the quality of the underlying businesses.

Cloudflare (NET) offers the broadest AI infrastructure angle, with its network positioned as a neutral layer between users, applications, and agents.

Okta (OKTA) and CyberArk (CYBR) play the identity and privileged access theme—probably the layer most directly affected by the proliferation of AI agents that need granular, auditable permissions.

Zscaler (ZS) presents the most interesting valuation case in relative terms. The stock has fallen roughly 36–50% from highs, trading around $130–140 when the analyst consensus points to $200–230 price targets. The fundamentals remain solid: in its fiscal Q3 2026 (ended April), Zscaler reported 25% year-over-year revenue growth to $850.5 million, with ARR reaching $3.525 billion (also +25% YoY). Management describes the company as “the cybersecurity platform for the AI era,” with its inline Zero Trust architecture specifically designed to secure agentic workflows at scale. The primary risk is decelerating organic growth and intensifying competition from PANW—real risks that deserve monitoring rather than dismissal.

At Poetic Street Capital, we have initiated a position in Zscaler representing approximately 3% of the portfolio, with an average entry price around $130 per share. This position reflects our belief that AI adoption is likely to increase—not decrease—the importance of secure access, identity-aware networking, and zero-trust architectures. We do not view the position as complete; should broader market volatility or company-specific concerns create materially lower prices in the future, we would be open to increasing our exposure.

The Historical Pattern Is Clear

The internet created network security. The cloud created cloud security. Mobile created mobile security.

AI is creating AI security.

The market’s attention remains heavily concentrated on direct AI winners: GPU manufacturers, hyperscalers, and foundation model developers. Yet history suggests that every major technological shift creates an entire ecosystem of secondary beneficiaries.

The Meta breach wasn’t evidence that cybersecurity is becoming less important.

It may have been one of the first visible signs that cybersecurity is becoming more important than ever.

Investors should spend less time asking whether AI replaces cybersecurity.

A better question is: Who gets paid to secure the billions of AI agents that corporations are about to deploy?

Disclosure: Poetic Street Capital currently holds a position in Zscaler (ZS). This article reflects personal opinions and is provided for informational purposes only. It should not be considered investment advice. Investors should conduct their own research before making investment decisions.

The philosophy was straightforward: if the position drifted from 8%, I’d rebalance. No rotation, no market timing.

Why I started questioning IEMA

The MSCI Emerging Markets index has become heavily concentrated in semiconductors — TSMC alone sits at ~11.5% of IEMA, with Samsung and SK Hynix adding another ~7%. People talk about the S&P 500 being over-concentrated in the Magnificent 7. Emerging Markets has its own version of that problem, just in semis and specifically in Taiwan and South Korea.

I believe semiconductors are overheated. Yes, AI demand for chips is real and massive — but the valuations have run far ahead of fundamentals.

The first rotation: IEMA → CBUK (partial)

A few months ago I rotated 3% of my portfolio from IEMA into CBUK (iShares MSCI China Tech), keeping 5% in IEMA. The goal was to reduce semi exposure while adding weight to beaten-down Chinese internet and AI names — Alibaba, Tencent, Baidu — that I believe are critical players in China’s AI race. China cannot afford to fall behind, and at some point the regulatory headwinds that have crushed these companies will ease.

The full rotation: everything into KWEB

More recently, I made a more aggressive call: I sold my entire IEMA position and rotated it fully into KWEB (KraneShares CSI China Internet ETF).

I initially considered staying in CBUK, which blends Chinese internet names with some semiconductor exposure. But looking at the relative performance chart between CBUK and KWEB, the two had tracked each other closely for a long time — and recently a gap had opened up in favor of CBUK. That divergence made KWEB look like the more attractive entry.

KWEB is pure Chinese internet: e-commerce, social media, digital services. Zero semiconductor weight. Exactly the exposure I want.

• Semiconductors: IEMA (19.6%), CBUK (9.2%), KWEB (0.0%)

• E-Commerce: IEMA (2.2%), CBUK (20.4%), KWEB (30.2%)

• Internet & Software: IEMA (3.3%), CBUK (22.2%), KWEB (20.5%)

The thesis

Chinese tech names go through cycles — periodic selloffs driven by regulatory fear and macro headwinds, followed by sharp recoveries. These companies are now central to China’s AI ambitions. That’s a structural tailwind that’s hard to ignore.

My play is KWEB vs. EEM: if Chinese internet rebounds while the semiconductor-heavy EM index gets hit by a semi correction, I rotate back from KWEB into IEMA and capture both moves.

Is this more of a value/tactical bet than my usual approach? Absolutely. I’m aware this sits awkwardly in an All-Weather framework. But the confluence of semi overvaluation and Chinese tech undervaluation felt too significant to ignore.

This is not investment advice. These are my personal decisions, shared for informational purposes only. Do your own research and consult a financial professional before making any investment decisions. I'll be posting updates on how this plays out — subscribe so you don't miss them.

Subscribe now

There’s a more rigorous way to think about it.

This piece lays out a systematic rotation strategy between gold and silver using the Gold/Silver Ratio as a signal — what it is, why it might work, what can break it, and why the implementation details matter more than the concept.

Two Decisions, Not One

The first thing to get right is separating two decisions that most investors collapse into one:

Decision 1 — Strategic allocation: How much of your total portfolio should live in precious metals permanently? This is a macro-level call about portfolio construction: hedge against monetary debasement, currency crises, geopolitical tail risk.

Decision 2 — Tactical composition: Within that allocation, what’s the split between gold and silver at any given time? This is a relative value call about two specific assets.

Conflating these is where most precious metals strategies go wrong. Someone buys silver because “it’s cheaper than gold and more upside potential,” which turns a structural allocation into a speculative bet. Or someone stays 100% gold forever, leaving a consistent historical anomaly completely untouched.

Keeping them separate is the foundation of what follows.

What the Gold/Silver Ratio Actually Is

The Gold/Silver Ratio tells you how many ounces of silver it takes to buy one ounce of gold. Today that number sits somewhere in the 80–90 range. Historically, it has oscillated between roughly 30 (silver euphoria) and 100+ (maximum fear, flight to gold).

But here’s the crucial insight: you’re not betting on silver going up. You’re betting on a statistical property — the tendency of this ratio to revert toward its historical range after reaching extremes.

That’s a fundamentally different thesis. And a much more defensible one.

The ratio has a plausible economic anchor: both metals share monetary history, similar mining economics, and correlated demand during financial stress. When the ratio hits 90, silver isn’t “cheap” in absolute terms — but it is historically cheap relative to gold. The question the strategy bets on is whether that relationship has some gravitational pull back toward equilibrium.

Centuries of data suggest it does. But — and this is important — not toward a fixed mean.

The Mechanics: A Rule-Based Rotation System

Here’s the basic framework:

Fixed allocation: 10% of total portfolio in precious metals, always. Rebalanced against the rest of the portfolio using current market values (liquid ETFs).

Internal rotation signal: The Gold/Silver Ratio.

• Ratio above 85 → rotate to silver (gold is historically expensive relative to silver)

• Ratio below 65 → rotate back to gold (silver has recovered toward equilibrium)

The 20-point band between triggers is deliberate. Narrow bands (buy silver at 75, buy gold at 70) generate too many rotations — tax events, spreads, noise. Wide bands sacrifice precision but add robustness. In a tax-sensitive environment, this matters enormously.

Why This Might Work

The strategy has three things going for it:

1. Historical persistence. The ratio has shown mean-reverting behavior across multiple economic regimes, currency systems, and decades. It’s not a perfect oscillator, but the extremes have consistently resolved.

2. Structural logic. Unlike many ratio strategies, there’s an actual economic reason for reversion: gold and silver are partially substitutable as monetary metals, their production costs are correlated, and institutional behavior during crises tends to drag them back into a historical relationship.

3. Behavioral edge. Most retail investors can’t sit patiently in silver when gold is ripping higher. A rules-based system removes that emotional override. The discipline is the alpha.

The Central Risk: Regime Change

Here is where intellectual honesty matters most.

The strategy doesn’t bet on silver. It bets on the stability of the ratio’s distribution. And there are two credible scenarios where that distribution could shift permanently:

Scenario A: Silver industrializes further. Silver’s industrial demand has been growing for decades — solar panels, EV components, electronics, data center infrastructure. If silver increasingly behaves like an industrial metal (copper-like) rather than a monetary metal (gold-like), its correlation with gold weakens. The ratio could structurally reset to a higher equilibrium, say 90–100, as silver’s monetary premium shrinks.

Scenario B: Gold remonetizes. Central banks have been accumulating gold at historically elevated rates since 2022. They are not buying silver. If the global de-dollarization trend continues and gold regains explicit monetary status in reserve frameworks, gold’s premium expands permanently. Again, the ratio’s center of gravity shifts higher — but this time because gold becomes more special, not because silver becomes less so.

In either case, the danger isn’t that the strategy produces one bad trade. The danger is that you wait for a reversion toward a “normal” of 60–70 that never comes because the new normal is 85.

The Binary Rotation Problem

This leads directly to the implementation question that matters most: should rotations be binary (100% gold ↔ 100% silver) or graduated?

Binary rotations are intellectually clean and emotionally satisfying. They’re also fragile.

Consider: you rotate to 100% silver at ratio 90. The ratio sits at 90 for seven years. Gold compounds at 12% annually. Silver compounds at 8%. You’ve been technically correct — the ratio is historically extreme — and you’ve systematically underperformed your alternative. In a portfolio where this 10% block exists primarily for defensive purposes, that’s not a rounding error.

A graduated approach is more robust:

• G/S ratio below 60: 100% gold 0% silver

• G/S ratio 60–70: 75% gold 25% silver

• G/S ratio 70–80: 50% gold 50% silver

• G/S ratio 80–90: 25% gold 75% silver

• G/S ratio above 90: 0% gold 100% silver

This preserves the mean-reversion thesis while maintaining meaningful exposure to gold — the primary monetary metal — even at extreme ratio readings. You still benefit if the ratio reverts. You’re not destroyed if it doesn’t.

The Costs Nobody Models

Before any backtesting looks convincing, run these numbers honestly:

Tax drag. In a high-bracket environment, each rotation triggers a taxable event. Capital gains taxes of 19–30% on realized gains can consume a substantial portion of the tactical alpha you’re trying to generate. A strategy that looks like +2.5% annual alpha pre-tax might be +0.8% post-tax — or negative, in certain rotation frequencies.

ETF friction. Bid-ask spreads on precious metals ETFs are small but not zero. Management fees (TER) compound quietly. Neither matters much in a buy-and-hold structure; both matter in a rotation strategy where you’re transacting on signal.

Opportunity cost of the block itself. If the 10% block exists for macro protection but you’re constantly optimizing its internal composition, you may be solving the wrong problem. The primary objective is portfolio insurance. The secondary objective — tactical alpha within the block — should never compromise the first.

The mental model: run a full simulation with realistic tax rates and transaction costs before assuming the strategy adds value net of all friction. Many strategies that are elegant in theory are marginal in practice.

A Note on Applying This Logic to BTC/ETH

The same framework is tempting to apply to Bitcoin and Ethereum, using the BTC/ETH ratio as a rotation signal.

It’s worth being direct: the analogy is structurally weaker, and the risks are categorically different.

Gold and silver share: centuries of monetary history, physical extraction dynamics, institutional ownership patterns, and a relatively stable economic relationship.

Bitcoin and Ethereum do not. Bitcoin increasingly resembles a digital monetary reserve — scarce, non-programmable, increasingly held by institutions and sovereigns. Ethereum increasingly resembles technological infrastructure — a programmable settlement layer whose value derives from usage and developer adoption. They are not two versions of the same asset. They are two different asset classes that happen to both be called “crypto.”

The ratio between them has no fundamental anchor. If Ethereum loses developer mindshare to competing chains in five years, the BTC/ETH ratio could double permanently — not because of a temporary anomaly, but because the underlying competitive landscape changed. There’s no geological mining relationship to hold it together. There’s no shared institutional demand dynamic.

The historical record is also insufficient. Gold/Silver has decades of consistent data across multiple economic regimes. BTC/ETH has roughly one decade of data, covering a single macro environment (near-zero rates, risk-on), one major technological transition (ETH’s merge to proof-of-stake), and a regulatory landscape that is still forming.

If you must apply rotation logic to crypto, do it with hard position limits (never below 30% in either asset), accept that you’re pattern-trading rather than mean-reversion investing, and size the position accordingly.

The Honest Portfolio Perspective

Here’s the framing that ties everything together.

If your precious metals block is 10% of a meaningful portfolio, even a well-executed tactical rotation generating +2–3% annual alpha on that block produces roughly +0.2–0.3% on total portfolio returns.

That’s not nothing. But it means the primary objective of this block — preserving wealth, providing uncorrelated returns during equity crises, hedging monetary debasement — should never be sacrificed for tactical optimization.

The rotation strategy makes sense as a secondary feature layered on a sound structural allocation. It becomes dangerous when it becomes the primary reason to own the block.

Own the metals because they belong in your portfolio. Rotate between them because the ratio offers a disciplined, historically grounded signal. Know the difference between the two.

Summary: What This Strategy Gets Right, and Where to Be Careful

Strengths:

• Separates strategic allocation from tactical composition — a distinction most investors miss

• Uses a ratio with genuine economic logic and centuries of data

• Rule-based system removes emotional decision-making at market extremes

• Wide trigger bands (85/65) reduce rotation frequency and friction

Modifications worth considering:

• Replace binary rotations with graduated bands (75/25 increments) to reduce regime-change risk

• Model tax drag explicitly before assuming positive net alpha

• Keep the block’s defensive function primary; rotation is secondary

Risks to monitor:

• Structural shift in silver’s identity (monetary → industrial) changing ratio equilibrium

• Accelerated gold remonetization by central banks widening the ratio permanently

• Extended periods where the ratio is “wrong” by historical standards but remains there for years

BTC/ETH verdict: The logic doesn’t transfer cleanly. Different asset types, insufficient history, technology risk. Proceed with smaller position limits and lower conviction.

This post represents a framework for thinking about precious metals allocation, not financial advice. All investment decisions carry risk and should be made in the context of your full financial situation.

Subscribe now

Introduction

Howard Hughes Holdings (NYSE: HHH) is a U.S. real estate company with a distinctive business model centered around the development and management of large-scale Master Planned Communities (MPCs) in high-growth markets.

Unlike traditional developers focused on individual projects, HHH acquires vast land positions and develops them over multi-decade periods, gradually creating integrated urban ecosystems that combine residential neighborhoods, office space, retail, hospitality, entertainment, and infrastructure.

This long-duration approach allows the company to continuously unlock value as infrastructure, population density, and commercial activity mature over time. In many ways, HHH operates more like a long-term land and infrastructure platform than a conventional real estate developer.

In recent years, the company has also begun transitioning toward a broader permanent-capital holding company model under the influence of Bill Ackman and Pershing Square.

Company History

Origins: Howard Hughes and Summa Corporation

During the 1940s–1970s, businessman and aviator Howard Hughes consolidated many of his assets — including airlines, television stations, casinos, hotels, and real estate — under Summa Corporation.

One of the most important acquisitions occurred in the 1950s, when Hughes purchased approximately 25,000 acres of desert land west of Las Vegas for around $3 per acre. That land eventually became Summerlin, now one of the most successful MPCs in the United States.

Transition Toward MPCs

Following Hughes’ death in 1976, many of the conglomerate’s operating businesses were sold or dismantled. Over time, management recognized that the combination of residential land development and ownership of adjacent commercial real estate produced exceptionally attractive economics.

The company increasingly focused on MPC development and was eventually renamed Howard Hughes Corporation.

GGP Bankruptcy and Spin-Off

HHH later became part of General Growth Properties (GGP), a major mall operator that entered bankruptcy during the Global Financial Crisis.

Bill Ackman played a major role in the restructuring process and ultimately separated Howard Hughes into an independent publicly traded company in 2010.

Seaport Spin-Off (2024)

In 2024, HHH spun off Seaport Entertainment, separating several entertainment-heavy Manhattan assets that required substantial capital and operational focus.

The transaction simplified the company structure and transformed HHH into a more focused MPC and real estate development platform.

Traditional Business Model

The company’s core business revolves around the development of Master Planned Communities.

This process typically follows several stages:

1. Land Acquisition

HHH acquires large strategic land positions in markets expected to experience long-term population and economic growth.

2. Master Planning and Zoning

The company designs long-term integrated communities including residential districts, retail, office space, schools, parks, and transportation infrastructure.

3. Infrastructure Development

HHH invests heavily in roads, utilities, landscaping, drainage systems, and other foundational infrastructure.

4. Residential Land Sales

Once infrastructure is completed, the company sells finished lots to homebuilders, who then construct and sell homes to end customers.

5. Development of Income-Producing Assets

In parallel, HHH retains ownership of selected commercial, multifamily, office, and retail assets inside its MPCs.

These properties generate recurring Net Operating Income (NOI) and create long-term recurring cash flows.

6. Optional Monetization

From time to time, mature assets may be sold to recycle capital into new developments or debt reduction.

Valuation Framework

Valuing Howard Hughes Holdings requires looking beyond quarterly earnings and focusing primarily on Net Asset Value (NAV).

Net Asset Value (NAV)

Many analysts believe the market value of HHH significantly understates the value of its land bank, stabilized assets, and long-term development pipeline.

MPC Economics

Several MPCs — particularly Summerlin and Bridgeland — continue generating record land sales and strong operating margins.

Free Cash Flow Evolution

As multifamily and office developments stabilize, recurring cash flows increasingly cover corporate overhead and debt servicing requirements.

Over time, this could reduce the company’s dependence on cyclical land sales.

Key Assets

A. The Woodlands (Texas)

The Woodlands is HHH’s most mature and stabilized MPC, with a population of roughly 123,000 residents.

Most of the developable land has already been monetized, making the asset function more like a stabilized cash-flow platform.

Main Assets

• Hughes Landing office complexes

• Luxury multifamily assets

• Retail centers

• Proximity to major corporate campuses including Occidental Petroleum and Exxon-related infrastructure

Cash Flow Generation

The Woodlands generates approximately $200 million annually in recurring rental revenue across office, residential, and retail segments.

The region also benefits from long-term corporate migration trends into Texas.

B. Summerlin (Las Vegas)

Summerlin is arguably HHH’s flagship MPC.

Built on the original desert land acquired by Howard Hughes decades ago, it has evolved into a major urban ecosystem with approximately 127,000 residents.

Asset Profile

• Downtown Summerlin retail district

• High-end residential communities

• Modern office developments

• Significant remaining land inventory

Economics

Summerlin produces extremely attractive land sale margins, with some transactions reportedly generating cash margins approaching 80%.

As remaining premium land inventory declines, pricing power may continue increasing.

Many investors view Summerlin as a near-monopoly luxury development platform in Las Vegas.

C. Columbia (Maryland)

Located between Washington D.C. and Baltimore, Columbia is a mature MPC with limited future expansion capacity.

The asset base is heavily oriented toward medical-office properties and multifamily real estate.

The region benefits from strong long-term occupancy trends driven by nearby healthcare infrastructure.

D. Teravalis (Arizona)

Formerly known as Douglas Ranch, Teravalis represents HHH’s largest long-term development opportunity.

The project could eventually support more than 100,000 residents over several decades.

Early Economics

Initial land sales have already begun at attractive pricing levels.

Key Risk: Water Rights

The primary uncertainty surrounding Teravalis involves Arizona water regulation and long-term water availability.

Although management is pursuing mitigation strategies, including recycling and optimization technologies, water access remains one of the most important risks to the long-term thesis.

E. Ward Village (Hawaii)

Ward Village is a luxury mixed-use development located in Honolulu.

Unlike HHH’s continental MPCs, Ward Village focuses primarily on high-end vertical residential development.

Competitive Advantage

The project benefits from an extremely limited supply of premium oceanfront development opportunities.

Economics

The development has already generated billions of dollars in cumulative sales and still maintains a substantial future pipeline.

Additionally, HHH retains ownership of selected retail assets within the district, generating recurring commercial income.

Sum-of-the-Parts Valuation

A simplified sum-of-the-parts framework suggests that HHH’s intrinsic value may materially exceed its public market valuation.

Stabilized Operating Assets

Estimated values for stabilized office, multifamily, and retail assets collectively imply several billion dollars of gross asset value.

After subtracting associated debt, these assets alone may justify a significant portion of the company’s market capitalization.

Land Bank Value

HHH owns one of the largest strategic land banks among publicly traded U.S. developers.

Given current absorption rates and development timelines, some analysts estimate that HHH possesses enough inventory to continue monetizing land for multiple decades.

Estimating the present value of this land bank is inherently difficult due to the extremely long duration involved.

Hawaii Development Pipeline

Ward Village alone could contribute several hundred million dollars of additional net value over time.

Intrinsic Value Estimate

Combining:

• stabilized operating assets,

• the land bank,

• Hawaii developments,

• cash balances,

• and net debt adjustments,

many investors conclude that HHH’s intrinsic value likely exceeds the current share price by a substantial margin.

Some estimates place conservative NAV around $90 per share, while management has previously referenced figures closer to $118–$120 per share.

The ultimate value could be significantly higher if Teravalis successfully replicates the historical trajectory of Summerlin or The Woodlands.

Bill Ackman and the “Baby Berkshire” Thesis

Through Pershing Square and affiliated vehicles, Bill Ackman now controls approximately 46–47% of HHH.

Ackman has publicly expressed interest in transforming HHH from a traditional real estate developer into a permanent-capital holding company inspired by Berkshire Hathaway.

The Vantage Acquisition

In late 2025, HHH announced the acquisition of Vantage Group Holdings for approximately $2.1 billion.

The strategic rationale closely resembles Berkshire’s historical model:

• generate insurance float,

• obtain low-cost capital,

• reinvest excess capital into long-duration investments.

Meanwhile, the MPC business would continue operating as a recurring cash-generation engine.

This raises the possibility that HHH could evolve into a hybrid structure combining:

• real estate,

• insurance,

• infrastructure-like cash flows,

• and eventually public equity investments.

Sources of Uncertainty

Despite the long-term upside narrative, several major risks and uncertainties remain.

Structural Complexity

Even as a pure real estate company, HHH was difficult for the market to value.

The addition of insurance operations and a potential holding-company structure increases complexity substantially.

Execution Risk

The acquisition of Vantage introduces HHH into a new industry outside its traditional expertise.

Insurance businesses only create value if underwriting discipline remains strong.

Weak Recent Financial Results

Recent earnings reports have disappointed investors, with declines in:

• net income,

• quarterly profitability,

• and adjusted operating cash flow.

These results have reinforced market skepticism.

Governance Concerns

Pershing Square’s growing influence creates potential conflicts between minority shareholders and Ackman-controlled entities.

One of the biggest concerns is whether future value creation remains inside HHH or migrates toward Pershing-related vehicles.

Pershing Square USA (PSUS)

In 2024, Ackman announced plans to launch Pershing Square USA (PSUS), a publicly traded closed-end investment vehicle.

The original ambition was extremely large, initially targeting one of the biggest investment fund launches in history.

However, institutional demand proved weaker than expected, leading to a temporary cancellation and redesign of the offering.

The vehicle was eventually relaunched in 2026 with a revised structure.

Why PSUS Matters for HHH Investors

The existence of PSUS creates an important strategic question:

Where Will the Compounding Engine Ultimately Sit?

Several future structures are possible:

Scenario A — HHH Becomes the Compounder

HHH internally retains:

• MPC cash flows,

• insurance float,

• and a growing investment portfolio.

In this scenario, HHH could evolve toward a Berkshire-like structure.

Scenario B — HHH Becomes the Capital Base

Alternatively, HHH could function primarily as:

• an asset-heavy cash generator,

• while Pershing entities capture the higher-multiple asset-light investment management economics.

This would resemble the broader Brookfield ecosystem:

• operating assets remain inside capital-heavy vehicles,

• while the premium valuation migrates toward the management platform.

In such a structure, HHH could still remain valuable, but the long-term upside profile for minority shareholders would likely be lower.

Potential Privatization Risk

One of the most discussed risks among investors is the possibility that Ackman eventually privatizes HHH.

If Ackman believes intrinsic value is materially above market price, acquiring the remaining shares could become economically attractive.

This creates a paradoxical situation for minority investors:

• a low share price may increase frustration in the short term,

• but it could also increase the probability of a take-private transaction at a meaningful premium.

Since Pershing previously purchased shares in much higher price ranges, many investors believe any future acquisition would likely need to occur at prices materially above current trading levels.

Berkshire vs Brookfield: The Core Strategic Debate

At the center of the HHH thesis lies one critical question:

Will HHH Become More Like Berkshire Hathaway or Brookfield?

Berkshire Hathaway Model

Under the Berkshire structure:

• everything remains under one roof,

• operating businesses,

• insurance,

• investments,

• and capital allocation.

The shareholder participates directly in the entire compounding engine.

Brookfield Model

Brookfield took a different path:

• management businesses received premium valuations,

• while capital-heavy assets remained in separate vehicles.

This structure often produces superior market multiples for the asset-light management entities.

Some investors fear Ackman could pursue a similar architecture:

• keeping MPCs and insurance inside HHH,

• while the most attractive investment-management economics migrate toward Pershing-related entities.

In that case, HHH could evolve into:

• a financing vehicle,

• a utility-like compounder,

• or a hybrid REIT/insurance platform.

Such an outcome would not necessarily be bad, but it would fundamentally change the investment thesis.

Future Scenarios

Possible long-term outcomes for HHH include:

• Berkshire-lite: Internal reinvestment engine combining MPCs, float, and investments

• Markel-like structure: Conservative compounder with operating businesses and investments

• Brookfield-style ecosystem: Asset-heavy HHH + asset-light Pershing platform

• Utility-like vehicle: Stable cash flows and dividends with lower growth

• Privatization: Ackman acquires remaining shares

• Permanent value trap: NAV grows but market discount persists

The ultimate shareholder outcome depends less on whether HHH owns valuable assets — and more on where future compounding economics ultimately reside.

Conclusion

Despite the complexity and uncertainty, HHH appears to offer a highly asymmetric risk/reward profile at current prices.

Even under relatively conservative assumptions, the underlying value of:

• MPCs,

• stabilized assets,

• land inventory,

• and future development rights,

may materially exceed the company’s public market valuation.

The central question is no longer simply whether HHH is undervalued.

Instead, the key issue is whether minority shareholders will fully participate in the long-term compounding engine that Bill Ackman is attempting to build.

At current prices, HHH may represent:

• a discounted real-asset platform,

• a future Berkshire-style compounder,

• or a utility-like permanent-capital vehicle.

Each outcome implies a very different long-term return profile.

We currently hold an approximate 2.6% portfolio allocation to HHH at an average purchase price near $63.6 per share and may increase the position during periods of additional weakness.

This document represents personal analysis and should not be considered financial advice or a recommendation to buy or sell securities.

Future updates and follow-up analysis on Howard Hughes Holdings will continue to be published as the story evolves.