"Microsoft removes two data centers" sparks heated discussion, Wall Street: Is the initial reaction of US stocks to DeepSeek incorrect?

Recently, Microsoft has canceled two data center projects, pouring cold water on the booming AI investment trend.

According to a report released by Michael Elias of TD Cowen, Microsoft has terminated several leasing agreements with multiple private data center operators, involving a total power capacity of hundreds of megawatts.

TD Cowen believes that these actions indicate that Microsoft may be facing a "supply surplus" in data centers. Microsoft's actions may just be the beginning. Given that Microsoft has the largest projected capital expenditure growth curve in Mag7, if it takes the lead in canceling leases and generally breaks the narrative of "sky-high capital expenditures," other companies are likely to follow suit quickly.

The market can't help but ask: Was the initial reaction of the U.S. stock market to DeepSeek incorrect? Is the AI capital expenditure feast that has driven the U.S. stock market up nearing its end?

About a month ago, when Wall Street first realized that DeepSeek was cheaper and more efficient, posing a challenge to the AI narrative, the market briefly panicked, and NVIDIA's stock price plummeted. Currently, competition in the AI field is becoming increasingly fierce, the pace of technological iteration is accelerating, and market expectations for AI prospects are becoming more cautious.

Cutting two data centers! Microsoft's "change of heart" has long been foreshadowed

According to the tech self-media “Consensus Crusher”, Microsoft has cut the Wisconsin Kenosha and Georgia Atlanta data centers. The former was due to Microsoft's non-participation in the Stargate project, transferring Wisconsin Kenosha to the Stargate project, with plans delayed by 3 years; the latter was due to overestimating the demand around Atlanta, with a delay of 1.5 years In addition, Microsoft has also suspended the conversion of negotiated and signed Statements of Qualifications (SOQ) into lease agreements and shifted international spending to the United States.

In fact, Microsoft's "change of heart" was not without warning. TD Cowen had previously discovered through channel checks that Microsoft had:

• Abandoned multiple 100MW deals that were in early/mid-stage negotiations in several markets.
• Allowed over 1GW of larger land area Letters of Intent (LOI) to expire.
• Given up at least five signed plots located in multiple first-tier markets.

TD Cowen pointed out at the time that the scale of Microsoft's abandonment of potential data center capacity and land acquisitions indicated that the primary demand signals Microsoft initially responded to had disappeared, and its interest in capacity may have shifted in relation to OpenAI.

Cooling Demand for OpenAI, Capital Expenditure Bubble Burst?

Elias noted that Microsoft was the most active capacity lessee in the first half of 2023 and 2024, during which it was procuring capacity based on forecasts that included incremental workloads from OpenAI. However, Microsoft's decision to suspend the construction of a data center in Wisconsin (which previous channel checks indicated was intended to support OpenAI) suggests that it may have procured excess capacity, particularly in areas where that capacity cannot substitute for the cloud.

In other words, Microsoft's previous aggressive expansion of data centers was likely based on high expectations for OpenAI's demand. But it now appears that this expectation may have been overly optimistic.

TD Cowen delicately interpreted the current situation as "oversupply." But a more straightforward interpretation is that the capital expenditure bubble, which was originally expected to inject up to $500 billion in new growth capital within just a few years to keep up with the exponential AI demand curve, may have burst.

It is worth noting that about a month ago, when Wall Street first realized that DeepSeek was cheaper and more efficient, posing a challenge to the AI narrative, the market briefly panicked, and NVIDIA's stock price plummeted. Although the market quickly rebounded afterward, the entry of DeepSeek undoubtedly added variables to the AI competition landscape.

Currently, competition in the AI field is becoming increasingly fierce, the pace of technological iteration is accelerating, and market expectations for AI prospects are becoming more cautious