Is Meta's computing power "flipping the table," signaling the need to hit the brakes on the golden age of AI chips?

Waking up, Meta's Zuckerberg did something significant: announced that Meta will officially sell its own AI computing cloud services.

As soon as the news was released, the chip sector in the US stock market immediately faced a fierce sell-off:

• Philadelphia Semiconductor Index: plunged 6.27% in a single day

• Micron: plummeted 10.57%

• SanDisk: plummeted 10.62%

• Intel: plummeted 9.03%

• Corning: plummeted 13%

A computing storm initiated by Meta is drawing the entire AI hardware industry chain into a heated debate about the "demand ceiling."

I. Meta's "Turnaround": From Burning Money to Selling Water

To understand why the market reacted so violently, we need to look at what Meta has been doing in recent years.

Over the past two years, Meta has been one of the most aggressive "buyers" in the AI arms race. Spending between $125 billion and $145 billion annually on capital expenditures, it has been frantically purchasing GPUs, networking equipment, optical modules, power, and cooling facilities—primarily to catch up with OpenAI and Anthropic in the large model race.

The problem is, the money has been spent, but the models have not been produced. Meta's own large models have consistently lagged behind major competitors in performance, resulting in a large amount of deployed computing power being idle or operating inefficiently.

To put it metaphorically: this is a marathon in the AI era, where Meta was desperately trying to catch up on the track, only to find itself falling further behind the competitors. So, it sat by the roadside and began selling water to passersby.

At the moment it sat down, it became a "computing power seller."

Specific data can better illustrate how substantial its resources are:

• By the end of 2025, Meta's AI computing power is equivalent to about 2.5 million H100s, totaling about 2GW of power scale.

• In 2026, Meta's capital expenditure guidance is $135 billion, corresponding to an additional 2-3GW of computing power.

• Rough calculations suggest that by the end of 2026, Meta's total available computing power may approach 5GW.

What does this mean? It implies that Meta's AI computing capability may have surpassed that of any tech company other than the hyperscale cloud providers. Moreover, it cannot use it all itself.

Therefore, the logic of renting out idle computing power makes sense—having assets sitting there unused is a waste, and monetizing them can at least make the books look better.

II. Why Did Chip Stocks Drop? Two Voices in the Market

Regarding Meta's sale of computing power, there are currently two different interpretations in the market.

Interpretation One: Bearish—The "Computing Power Surplus" Signal Has Been Fired

The bearish logic chain is straightforward:

Meta has been one of the largest buyers of AI chips in the past two years. Now it is starting to rent out excess computing power—this indicates that it no longer needs to buy more.

If even a player of Meta's caliber is beginning to realize that it has "overbought," what about other manufacturers? Is the large model arms race shifting from "full sprint" to "slamming the brakes"?

A further extrapolation is: once the consensus of "computing power surplus" is formed, the expected incremental demand for the entire AI hardware market will be significantly downgraded. Core players in the supply chain like Samsung, TSMC, Micron, and NVIDIA may face risks of slowing order growth or even order cuts. The hardware industry chain story, which has been thriving for over two years, may begin to discount.

Interpretation Two: Bullish—"Selling Water" Is to Continue Running the Marathon

The bullish rebuttal is equally compelling:

The GPUs that Meta spent hundreds of billions of dollars on in the past two years are already sunk costs. Now, activating and monetizing these idle assets is not abandoning competition, but a return to commercial rationality.

If renting out computing power can indeed generate revenue, Meta's subsequent procurement of GPUs, networking equipment, and optical modules will be even more confident—because with the ability to recover funds, burning money can be more sustainable. Spending money on equipment → renting out idle computing power to recover funds → using the recovered funds to buy more equipment is actually a positive cycle, not a zero-sum game.

III. Who Is the Winner? The Capital Market's Answer Is Clear

The two viewpoints are currently in fierce contention, and the volatility of chip stocks is unlikely to calm down in the short term. However, one thing is certain—the capital market is rewarding Meta's decision to "sell water" with real money.

On the trading day following the announcement, Meta's own stock price did not plummet alongside chip stocks. On the contrary, it surged nearly 9%. This indicates that investors' attitudes are clear: regardless of how chip stocks fall, this move by Meta is a positive for itself.

Why is this the case?

The core issue is not "how much money can be made from renting out computing power." Because even if Meta rents out all its surplus computing power, it is hard to say how much incremental net profit it can bring in the short term. Perhaps the initial figure is only in the range of $2 billion to $3 billion, which is not critical for a company with annual revenues in the hundreds of billions.

What the market really cares about is the change in attitude.

In recent years, Wall Street's biggest anxiety about Meta was not its GPU purchases, but its "limitless burning of money." Annual capital expenditures of $125 billion to $145 billion felt like an ever-deepening pit, raising investors' doubts about return on investment (ROI).

Against this backdrop, Zuckerberg's willingness to monetize surplus computing power—even if the amount is not large—conveys the signal that management is beginning to care about capital efficiency, and the phase of "crazy money burning" may be nearing its peak.

This is the signal Wall Street has been waiting for. Therefore, even as chip stocks plummet, Meta's own stock price rises against the trend. The capital is not rewarding the computing power rental business itself, but the strategic shift from "unlimited arms race" to "return to commercial rationality."

IV. The Real "Black Swan" May Not Have Appeared Yet

Although chip stocks have dropped sharply, it may still be too early to say that the AI hardware bull market has ended.

The biggest uncertainty currently lies not only in Meta's rental of computing power but also in whether other giants will follow suit.

Meta is the first leading tech company to publicly declare, "We have overbought computing power and need to rent it out." What about Microsoft and Amazon? If these cloud giants, which are also aggressively purchasing AI infrastructure, cannot withstand the pressure from the capital market and announce a need for "rational investment" and cut capital expenditure plans—then that would be the moment that truly shakes the foundation of the entire AI hardware industry chain.

So far, Microsoft and Amazon have not released similar signals. However, if in the coming weeks or months, "cutting capex" becomes a collective action among tech giants, the value of AI chip hardware will be reassessed.

In other words, the current market is shifting from one question to another: the issue is not with Meta, but whether Meta will be the first domino to fall.

V. In Conclusion: Volatility Is Certain, Direction Is Still Uncertain

For investors focused on chip stocks, the current situation can be summarized in one sentence: uncertainty is extremely high, and volatility will intensify in the short term.

How significant is the impact of Meta's rental of computing power on chip demand? Is it an isolated event or the beginning of an industry trend? Will Microsoft and Amazon follow suit? These questions currently have no answers. Before the answers surface, the stock prices of the chip sector are likely to continue fluctuating.

For investors looking to participate in the volatility trading of chip stocks, the BIT (formerly Matrixport) platform provides a real trading channel for Nasdaq-listed chip stocks—covering core targets such as Micron, Intel, AMD, and NVIDIA. What you buy are the actual stocks registered within the DTC system, not tokenized certificates, allowing for trading at any time and transfer through ACATS.

With BIT's USDT 24/7 near-instant deposit channel, even if you are in the Asian time zone, you can complete fund transfers before the US stock market opens, ensuring you do not miss any important market windows.

The story of AI hardware is not over, but the script is being rewritten.

Disclaimer: This article is for informational reference only and does not constitute investment advice. The market interpretations and data in the text are compiled based on publicly available information and may be subject to delays or deviations. The volatility risk of US stocks and the chip sector is high, and past performance does not represent future returns. The specific terms of related trading services are subject to the official explanations of the BIT platform, and users in different regions should confirm compliance on their own. Investment carries risks; please consult a professional advisor before making decisions and bear the corresponding risks and consequences.