Keep the glittering casket and give back the pearls: Meta's computing power business

There is an ancient Chinese story. A merchant from the State of Chu went to the State of Zheng to sell pearls. He made a case with magnolia wood, scented it with cinnamon and pepper, adorned it with pearls and jade, decorated it with roses, and inlaid it with emeralds. A man from Zheng bought the case and returned the pearls to the merchant.

When Han Feizi wrote this fable, he intended to satirize those who neglect the essentials and pursue the trivial. However, if we change the perspective and look at it from the Chu merchant's point of view, it is also an unintentional business strategy: he originally only wanted to sell the pearls, but unexpectedly found that the case he made was more valuable than the pearls in others' eyes.

On July 1, 2026, a report from Bloomberg gave this ancient fable a modern version. Meta is establishing a cloud infrastructure business unit to sell AI computing power and models to external customers. This business is tentatively named Metamate, with the goal of generating an annual revenue of at least $10 billion to $15 billion by the end of 2027.

In other words, the nearly $200 billion that Meta burned in the past two years was originally intended to build a large enough computing power pool for its advertising business and the Llama model. As a result, the pool was too large and mostly idle most of the time. So it decided to package the idle computing power into a commodity and sell it to companies that want to train AI but can't buy GPUs. It originally only wanted to raise a fish, but unexpectedly dug out a lake where it could sell water. The case has become so valuable that now some people only want the case, not the pearls.

What are Meta's pearls? They are advertisements. They contribute 98% of the company's revenue. The open - source Llama series models are the most eye - catching ornaments on Meta's case. They have been downloaded over a hundred million times but generate no revenue at all, belonging to pure craftsmanship display.

Would a rational businessman spend all his budget on decorating the case for the pearls? Meta's answer is: yes, and it has already done so. The capital expenditure in 2025 was nearly $100 billion, and the guidance for 2026 is set at $125 billion to $145 billion. The free cash flow has been pushed to the lowest point since 2014.

The case is getting more and more exquisite, but the price of the pearls is not rising accordingly. Investors keep asking the same question: When will this case start to make money on its own?

Metamate is the answer to this question. It is not the best answer, but it is the only answer that can be given without shutting down the decoration workshop.

Meta's logic is very simple: it has built one of the world's largest AI computing power pools, and the GPUs have a lot of idle cycles after supporting internal business. Instead of letting the expensive heat idle in the data center, it is better to rent them out by the hour. Just like a cake shop that built an industrial - grade baking workshop to bake one cake a day for itself. Now it finds that the workshop is mostly idle, so it starts to hang a sign at the door: Accepting baking services on behalf of others.

But next to this baking workshop are three time - honored brands that have been in business for more than ten years. AWS, Microsoft Azure, and Google Cloud together account for more than two - thirds of the global cloud market share. They have a mature customer system, perfect compliance certification, an experienced direct sales team, and a cooperation relationship with the government and large enterprises that has lasted for more than ten years.

And Meta, a social network company with the motto of "Move fast and break things", now has to enter the doors of these large customers and say: Hello, please entrust your most sensitive AI training tasks to me.

This scene itself is worth pondering. Andrew Bosworth, as the CTO, has taken on this task. Meta is assembling a sales team internally, trying to use the open - source Llama as a foot - in - the - door tool. The logic is self - consistent: AWS and Azure sell computing power plus their own closed - source models, or computing power plus their partners' closed - source models. Meta sells computing power plus the open - source Llama. For European and Asian - Pacific customers who are sensitive to data privacy and have strict compliance requirements, open - source means controllable, auditable, and capable of private deployment, without having to worry about being locked in by a closed - source model of a single supplier.

But open - source also means another thing: Llama itself is free. Meta can only make money from pure computing power leasing. This requires extremely high scale efficiency and extremely low marginal cost. Meta's confidence comes from its self - developed MTIA chips and years of data center operation experience, but confidence does not equal a moat.

The most ingenious part is hidden in the timeline.

Meta has set a goal for Metamate to generate an annual revenue of $10 billion to $15 billion by the end of 2027. For a company with an annual revenue of over $200 billion, this figure is not large. What really matters is that this revenue will prove to Wall Street that AI can become a profitable business, rather than just a bottomless cost pit.

A revenue of $15 billion itself doesn't change much. What's important is the narrative shift it brings. If Metamate succeeds, Meta will transform from "a social network company supported by advertising" to "a technology infrastructure company with an open - source AI ecosystem". There is an entire trillion - dollar valuation ceiling between these two narratives.

On the day the news was announced, Meta's stock price rose immediately. Investors are using real money to express a judgment: When the world's largest AI computing power buyer starts to become a seller, the underlying logic of the entire industry is quietly changing. Infrastructure is no longer the exclusive means of production of a single company; it is becoming a tradable commodity. Just like an electric power company that originally only supplied power to its own factory, and later found that the whole city was queuing up for electricity.

When Han Feizi wrote the story of "Buying the casket and returning the pearls", he probably didn't expect that this story could be used to explain the strategic transformation of a Silicon Valley giant. But the choice of that man from Zheng may not seem ridiculous today. He valued the craftsmanship, materials, and design of the case, and the value of these things is independent of the pearls. When he bought the case, he bought the underlying construction that others had spent a lot of effort on.

Meta has found that its ability to make cases has become strong enough to be sold separately. This is a rare encounter that could only happen in the AI era: the pearls are not valuable, but the case for the pearls has become a new type of asset.

This article is from the WeChat official account "Silicon - based Starlight", author: Xingmang. Republished by 36Kr with permission.