AI computing power frenzy cannot hide the "fear of heights" in capital expenditure: Meta and Anthropic announce new moves one after another, and the market begins to reassess AI deals?
```
The AI hardware sector has adjusted for two consecutive days, but what has truly attracted market attention is not the chip companies themselves, but the latest moves by two AI large model companies.
On Wednesday, it was reported that Meta is exploring the commercialization of surplus AI computing power. A day later, media reported that Anthropic is discussing cooperating with Samsung Electronics to jointly develop AI chips and is considering using Samsung's 2-nanometer process for manufacturing.
The two pieces of news seem unrelated, but together they touch on the most sensitive topic in the current AI industry chain—whether the AI capital expenditure, which has expanded rapidly for two years, is entering a new stage?
The market took the lead in repricing. U.S. chip stocks have generally plummeted over the past two days. The Philadelphia Semiconductor Index (SOX) dropped more than 10% cumulatively on Wednesday and Thursday, marking the largest two-day drop in a month. The semiconductor equipment sector, which is most sensitive to capital expenditure cycles, led the decline: Teradyne (TER), Entegris (ENTG), KLA (KLAC), Applied Materials (AMAT), and Lam Research (LRCX) at one point each dropped over 10% during Thursday's trading. European chip giant ASML's U.S. shares (ASML) once fell over 5% on Thursday.
In contrast, many institutions believe the two pieces of news are more like catalysts for the market to reassess the AI investment logic rather than a fundamental reversal in the industry's prosperity. What the market is actually trading is not "whether AI demand has peaked," but that the AI industry is shifting from "competing for capital expenditure" to "competing for capital efficiency"—a new stage.
What the market truly worries about is not Anthropic making chips, but the changing logic of AI capital expenditure
Over the past two years, the AI hardware sector has soared. The core logic has remained almost unchanged: rapid iterations of AI models continue to drive exploding demand for computing power, GPUs are persistently in short supply, tech giants keep raising capital expenditures, which in turn drives demand for GPUs, high-bandwidth memory (HBM), high-speed networks, advanced packaging, and semiconductor equipment, forming an unprecedented "AI capital expenditure super cycle."
This logic not only pushed Nvidia to become the company with the highest market value in the world, but also made equipment manufacturers like Applied Materials, Lam Research, Dutch ASML, KLA, as well as storage vendors like Micron and SanDisk the biggest winners in the capital market.
However, the two pieces of news that broke out over two days this week have prompted the market to seriously discuss: If the AI industry starts to focus more on capital efficiency rather than simply expanding investment, will the current capital expenditure super cycle enter a new stage?
On Wednesday, reports said Meta is planning to build an AI cloud computing business and may open AI models deployed on Meta's infrastructure to external customers, or directly lease surplus AI computing power to commercialize the return on its tens of billions of dollars of AI infrastructure investment.
Shortly after, news broke on Thursday that Anthropic is discussing developing self-designed AI chips.
Individually, the two companies are taking different paths, but together, they point toward one change—AI companies are beginning to consider how to improve returns on existing infrastructure investments, not just keep expanding capital expenditure.
This change in expectations has led to a reevaluation of AI trading logic in the market.
Anthropic’s self-developed chip—does this mean AI companies are entering a "cost optimization era"?
Compared to the market's initial concern about "whether self-developed chips will reduce GPU purchases," the commercial logic behind Anthropic’s move is more noteworthy.
It is reported that Anthropic is discussing with Samsung Electronics the development of customized chips for AI training and inference, currently still in early stages.
If it moves forward, Anthropic will become another foundational model company, after Google, Amazon, Microsoft, and Meta, to work on self-developed AI chips.
This doesn’t mean abandoning Nvidia GPUs, but is a natural evolution of AI industry development.
Over the past two years, the focus of competition among large model companies has been who could get more GPUs and build more data centers. As models grow in scale, the costs of training and inference rise rapidly; reducing per-Token costs, improving computing utilization, and decreasing reliance on a single supplier have become new priorities.
ASICs designed for specific models can achieve better balance among performance, power consumption, and cost, which is exactly why Google’s TPU, Amazon’s Trainium, and Meta’s MTIA have continued to push ahead in recent years.
In this sense, Anthropic pursuing self-developed chips is more an important hallmark of the AI industry moving from "competing for investment" to "competing for efficiency," rather than reducing AI investment.
Meta and Anthropic—two different paths toward the same goal
Meta and Anthropic have adopted different strategies but their goals are highly aligned.
Meta hopes to generate income from temporarily idle AI computing power to improve returns on its tens of billions in capital expenditure; Anthropic aims to lower long-term computing costs via custom chips and strengthen its autonomy over infrastructure.
Whether it’s selling surplus compute or deploying ASICs, essentially it’s not about reducing AI investments, but seeking more sustainable AI business models.
However, for capital markets, these two pieces of news easily trigger another line of thinking: If AI companies start focusing on capital efficiency, will GPU purchases, cloud rentals, and new data center investments in the future still grow as rapidly as in the past two years?
As a result, the market is reassessing whether AI capital expenditure can maintain its previous "only increase, never decrease" expectations.
That’s why in the two days of adjustment, the biggest losses came not from model companies, but semiconductor equipment firms most closely tied to new capital expenditure. Compared to GPU and memory vendors, equipment orders more directly reflect investment plans of wafer and chip companies, making them most sensitive to changes in capital expenditure expectations.
Institutions: The market is more like repricing AI trades rather than denying the AI super cycle
Despite consecutive days of semiconductor stock declines, most institutions do not interpret the two items of news as cooling AI demand.
Regarding Meta, many analysts believe selling surplus compute is more about finding a commercial outlet for its massive AI capital expenditure, thus improving the sustainability of future investments in GPUs, network gear, data centers, and energy infrastructure, not cutting investments.
For Anthropic, institutions generally agree that self-developed chips match the long-term trends of AI large model companies. Even if more companies shift to ASICs, they will still rely on advanced process manufacturing, HBM, high-speed interconnects, advanced packaging, and data center construction. Demand for AI infrastructure will not disappear but may be redistributed among various segments.
More importantly, the penetration rate of AI applications is still quite low. Industry insiders point out that as inference demand keeps rising, Token consumption and computing needs for large models are far higher than previously expected. True maturity of AI infrastructure is still some way off.
Therefore, the market this week looks more like a phase of repricing for AI trades after historic rallies.
If the past two years of AI competition were about "who invests more," the signals sent by Meta and Anthropic indicate that AI is entering a new stage—competition is shifting to who can make every dollar of capital expenditure deliver higher returns.
For the market, this expectation transition is enough to be a catalyst for adjustment in the AI hardware sector; for the industry itself, it does not necessarily mean the end of the super cycle, but rather could signal AI infrastructure investment is entering a more mature, more business-focused stage.
Risk warning and disclaimerThe market has risks, investment requires caution. This article does not constitute personal investment advice, nor does it take into account individual users' specific investment goals, financial status, or needs. Users should consider whether any opinions, views, or conclusions in this article suit their own circumstances. You will bear sole responsibility for any investments made based on this article. ```