Author: Mr. Munger from the US Research

Yesterday, the entire market was collectively optimistic about AI infrastructure, but today, optical modules, storage, and equipment have been cut by 7% to 15%.

The trigger was a small article stating that "Meta wants to sell computing power."

In this article, I will analyze in depth why this logical chain is incorrect, along with two important flaws in most follow-the-trend analyses, and finally a counterexample and my conclusion.

My view is very straightforward: the market is trading a foolish syllogism today. You can believe anything, but not this small article.

The biggest mistake in today's trading is that it extrapolates a single company's business decisions directly into supply and demand signals for the entire industry.

Let’s break it down layer by layer.

1️⃣ The cause of the matter: a small article stating "Meta wants to sell computing power"

Bloomberg first hinted that Meta is planning a cloud business and may sell "excess AI computing power" externally, even selling raw compute directly. Reuters followed up with a key point: the plan is still in development, strategies may change, Meta declined to comment, and the news has not been independently verified.

With such an unconfirmed rumor, the market immediately deduced a logical chain, amplifying the public sentiment that does not understand the AI computing power industry:

Meta wants to sell computing power → Meta itself cannot use it all → AI training/inference demand is not good → AI computing power is in excess → The scarce rents originally belonging to Nebius and CoreWeave will be taken away → The AI hardware chain should decline.

Thus, you can see that almost all AI infrastructure stocks are falling.

Why do I think this is wrong? Read on.

2️⃣ The market is trading a foolish syllogism

The market is currently trading a deliberately rendered logical chain:

Meta wants to sell computing power = Meta cannot use it all = AI is not viable = Computing power is in excess = The entire industry chain should fall.

But the problem is whether Meta sells computing power is a question of how a single company's balance sheet is calculated; whereas whether "the world lacks AI computing power or not" is a completely different question.

Next, I will break down the two biggest flaws.

3️⃣ Flaw one: Meta selling computing power ≠ AI computing power excess, quite the opposite

If Meta is indeed selling computing power because of "oversupply and AI is failing," then the following actions do not align at all.

In April, Meta raised its 2026 CapEx guidance to $125 billion to $145 billion. Which company would continue to increase capital expenditure when anticipating a collapse in demand?

In April, Meta signed an AI cloud agreement with CoreWeave worth about $21 billion, locked until 2032, including Vera Rubin GPU capacity; in March, it signed a multi-year agreement with Nebius worth up to about $27 billion. Just three months ago, AI suddenly became in excess? Meta's management isn't that fickle, especially since large company's budgets are typically set a year in advance.

Recently, it was also reported that Google restricted Meta's use of Gemini due to Meta's demand exceeding what Google can supply. If demand has already reached a level that others cannot meet, how can you tell me that it is in excess? (I've analyzed this viewpoint in a previous post.)

4️⃣ Flaw two: Even if Neocloud's logic holds, why kill the "shovel sellers" along with it?

For CRWV and NBIS, I can barely understand the market's concerns: if Meta becomes a potential competitor from a super customer, it is indeed a discussion-worthy issue, provided you truly believe it will do so.

However, extending this logic to optical modules, interconnects, storage, and equipment makes no sense.

These companies are essentially "shovel sellers" within the AI infrastructure supply chain, and their relationship with Meta's sale of computing power is almost nonexistent.

The larger the AI cluster, the greater the need for high-speed interconnects, optical modules, and lasers. Renting out some of its computing power will not make these physical requirements disappear or reduce the demand for a single optical fiber in the entire cluster.

The same applies to storage.

Equipment manufacturers (like AMAT) are even more typical; they are at the very bottom of the entire supply chain. Whether Meta sells computing power has no direct transmission mechanism affecting whether wafer fabs continue to procure etching, deposition, CMP, and other equipment, which is thousands of miles apart.

5️⃣ The biggest flaw: Why did Oracle hardly decline?

This actually reveals a lot of issues.

If the market truly believed that "AI is done" and "computing power is in excess," then Oracle, as a player in AI cloud holding a large amount of GPU cloud resources, should drop even more, especially considering its negative FY free cash flow and CapEx guidance raised to nearly $100 billion.

Instead, Oracle remains stable, Neocloud crashes, and shovel sellers fall in tandem.

This combination in itself indicates that the market is not killing demand today; "computing power excess" seems more like a guise.

6️⃣ The real issue is demand-supply mismatch, not oversupply

This is what I consider the crux of the whole matter.

The overall market may be facing shortage while a specific company, at a specific moment, in a specific region, or with a specific generation of GPU may have temporary excess, which is not contradictory at all.

GPUs, electricity, data centers, networks are originally built several years in advance; while model training is project-based, API release timings may be delayed, and inference traffic grows non-linearly, leading to a natural mismatch.

As a result, it can lead to: a long-term strategic lack of computing power but short-term financial local resource gaps.

This is called Mismatch (demand-supply mismatch), not Oversupply .

Conclusion

Meta is not selling NVIDIA because "AI is failing."

On the contrary, a more reasonable explanation is: the inference business is highly profitable with strong demand, and Meta hopes to monetize the temporary excess capacity while improving its profit statement.

However, the market misinterpreted a corporate-level resource optimization as the entire AI industry entering a state of oversupply.

And I believe this is the most core yet easily overlooked misjudgment in today's collective decline of AI infrastructure.