How did a data center storm trigger a tech stock crash?
After the U.S. midterm elections, policy risks for data center construction rose. On November 5, the Nasdaq index plunged 5.5%, AMD fell 17.3%, and Nvidia fell 9.3%. An Anthropic-related data center project in Hubbard, Texas, faced obstacles in securing about $15 billion in loan financing. UBS estimates that about 30% of the planned new U.S. data center capacity for 2028 may be affected by construction restrictions. Citrini Research believes that the investment logic that drove the broad rally in AI stocks over the past three years has entered a turning point, while Fundstrat's Tom Lee believes the market has overpriced the policy shock.
Why read this
AI score 69
The article links data center policy risks, financing obstacles, and stock price divergence into a single transmission chain, making it possible to see the turning point in the investment logic of AI infrastructure.
Topics
Body · AI translation
In just two days, a dramatic reversal played out on Wall Street.
On November 4, US local time, the Dow Jones Industrial Average rose about 3%, the S&P 500 rose 1.6%, and the Nasdaq rose 1.4%. Microsoft and Amazon rose 6.7% and 6.1% respectively, as large technology stocks led the market to a rebound after the US midterm elections.
But even as the broader market rose, the AI infrastructure sector had already seen pronounced selling. Data center developer Hut 8 plunged 23.2%, cooling equipment maker Modine fell 21.4%, power and cooling equipment maker Vertiv fell 12.9%, and AI cloud service provider CoreWeave fell 9.9%.
A day later, the selling spread rapidly.
On November 5, the Nasdaq index plunged 5.5%, the S&P 500 fell 4.4%, and the Philadelphia Semiconductor Index fell about 12%. AMD fell 17.3%, Nvidia fell 9.3%, and memory chip and AI server-related stocks broadly suffered heavy losses.
In fact, this storm occurred one day before the broader market plunge. On November 4, the rise in large technology stocks masked the collective decline in the AI infrastructure sector; on November 5, market concerns about data center construction and future capital expenditure spread further, ultimately turning into a broad selloff in technology stocks.
This plunge occurred at a delicate moment.
After the US midterm elections, the Democratic Party gained control of both chambers of Congress and won important election victories in Texas, Georgia, and other key data center construction hubs. Local voters' opposition to data centers' electricity and water use and costs to residents began to translate into more concrete policy risks.
At the same time, Wall Street research firm Citrini Research published a new report, "The End of the Trade (as we know it)," arguing that the investment logic that drove the broad rise in AI-related stocks over the past three years has entered a turning point.
01. A $15 billion loan
The most direct trigger for this round of selling was investors reassessing the outlook for US data center construction.
Over the past few years, demand for data centers has continued to grow, and power, land, and grid connection resources have become increasingly scarce. In the eyes of the capital markets, these bottlenecks once constituted new investment opportunities. Companies with power resources, construction capabilities, and cooling technology all received higher valuations.
After the US midterm elections, investors worried that local governments might require data centers to bear more grid construction costs, clean energy investment, and community compensation, and that project approval times might also be extended.
UBS estimates that about 30% of the US data center capacity planned for addition by 2028 could be affected by potential construction restrictions. Goldman Sachs noted that states that may face policy constraints account for about 45% of planned US data center capacity, with Texas accounting for a significant share.
This risk exposure has already begun to appear in specific financing transactions.
On October 28, Bloomberg reported that financing for an Anthropic-related data center in Hubbard, Texas, had encountered obstacles.
A banking syndicate led by Morgan Stanley had already arranged about $15 billion in loans for the project. Because Texas suspended some new project construction and was assessing grid risks, the banks postponed a transaction originally planned to sell part of the debt to bond investors. As a result, the banks need to hold the related risk assets for longer than originally planned.
This $15 billion loan is not a bond issued directly by Anthropic, but in such a sensitive period, it formed a chain of panic transmission. Delays in data center construction could postpone subsequent equipment procurement and leasing revenue; if the financing arrangements cannot proceed as planned, banks and project investors will also need to bear funding costs again.
Citrini believes that the funding environment that has supported AI infrastructure construction in the past is changing. Large technology companies can still raise financing on the strength of relatively strong cash flow and credit quality, but emerging AI cloud service providers lacking guarantees from hyperscale cloud vendors may face greater financing pressure.
Previously, AI data center operators obtained financing through long-term computing power contracts and then used debt to buy GPUs and build computer rooms. Customers' future computing power procurement commitments constituted an important basis for financial institutions to evaluate projects.
Long-term contracts once regarded by the market as guarantees of future revenue are beginning to be tested by both financing conditions and actual construction progress.
02. The further upstream, the harder the fall?
The market reversal on November 4 and 5 also reflected a very special economic relationship within the AI industry.
On November 4, the stock prices of large US technology companies generally rose. Investors believed that if future data center expansion slows, Microsoft, Amazon, Google, and other companies might reduce capital expenditure and ease pressure on free cash flow. This might not be a bad thing.
For companies with mature cloud businesses and large existing computing power resources, slower construction of new capacity could also increase the scarcity value of existing resources.
But for data center developers, chip manufacturers, and equipment suppliers, if a technology giant cuts $10 billion in data center construction spending, the chip, server, optical communications, and power equipment orders originally corresponding to that $10 billion could be delayed or even trigger a chain reaction.
This also explains why AI infrastructure stocks had already begun to fall on November 4 while the broader market rose. A day later, the market further worried that if the construction slowdown continues, it might ultimately also limit the future service capacity and revenue growth of large cloud vendors. Panic and selling spread as a result.
Figure: Since the second half of 2025, the scale of remaining performance obligations at the four companies has expanded significantly, but Citrini points out that their quarterly growth rate has fallen to a relatively low level in recent years, indicating that the pace of expansion in contract backlog is slowing.
On November 5, CoreWeave's stock price fell about 10%, and Nebius fell about 15%. More noteworthy is that CoreWeave's five-year credit default swap spread rose to a record high, indicating that market concerns about its credit risk are also rising.
Subsequently, the pressure spread to the semiconductor and storage industries.
AMD fell 17.3% in a single day, and Nvidia fell 9.3%; SanDisk fell about 22%, and Micron and Western Digital both fell more than 17%.
Their business structures and profitability differ greatly, but there is one common problem they must face: from 2027 to 2028, can AI infrastructure investment still maintain the growth rate previously expected?
Power and cooling equipment companies were not spared either. GE Vernova fell 7.5% on November 5, after already falling about 9.3% the previous day.
Figure: GE Vernova's gas turbine manufacturing facility in Greenville, South Carolina, USA. The expansion of AI data centers has driven demand for power equipment, and changes in construction plans may also affect suppliers' future order expectations.
Capital markets have always traded the future, and the future, more than the present and the past, is better able to stir the market's sensitive nerves.
This round of adjustment has therefore also shown clear divergence: companies that rely more heavily on financing conditions and new construction projects are more vulnerable to shocks; enterprises with existing capacity, mature business cash flow, and diversified customers theoretically have stronger buffering capacity.
However, policy and construction risks are only one trigger. Previously high valuations, concentrated holdings in technology stocks, divergent corporate earnings, and macro fund deleveraging also jointly amplified the decline.
03. Wall Street divergence: Is the AI trade over, or is the market overreacting?
On November 4, Citrini Research proposed in its latest "State of the Themes" report that the most optimistic phase of the familiar approach to AI infrastructure investment may already be over.
This institution began building portfolios around long-term themes such as AI in 2023 and once achieved relatively high returns by holding companies benefiting from the technological revolution. But it now states that the current market environment is clearly different from three years ago.
Figure: Cumulative return performance of Citrini Research's portfolios and five major thematic strategies since inception. As of October 26, 2026, the Citrindex actual portfolio had a cumulative return of 257.2%, among which the AI thematic strategy once exceeded 500% before falling back significantly.
In 2023, valuations of companies such as Nvidia were relatively low, and AI investment had only just begun. Investors only needed to find companies that could benefit from growing computing demand to have a chance to share in the returns from the entire industry's expansion. The valuations of many stocks already incorporated optimistic forecasts for the market size in 2030.
Citrini estimates that the AI ecosystem's annualized revenue must reach at least $200 billion, while the annualized capital expenditure of large cloud vendors is approaching $1 trillion.
If capital expenditure continues to expand rapidly, the revenue of model companies and application companies will need to grow further to support the investment returns of the entire industrial chain.
Figure: Announcement amounts of computing procurement commitments by OpenAI and Anthropic from January 2023 to October 2026, as compiled by Citrini Research. Since the second half of 2025, the large computing contracts disclosed by the two companies have increased significantly, with OpenAI's highest single-month announced amount reaching $306.5 billion, and Anthropic's related commitments disclosed in October 2026 reaching as high as $120.5 billion.
But Wall Street has not formed a consensus judgment. Dan Ives, who has long been bullish on technology stocks, also believes that the regulatory and approval pressures facing data center construction may cause the growth rate of future capital expenditure to peak earlier. He remains bullish on AI demand, but believes that the adjustment in some semiconductor stocks may not yet be over.
Tom Lee of Fundstrat holds a relatively optimistic view. He believes that local governments need the tax revenue, jobs, and investment brought by data centers, and may ultimately reach a compromise with technology companies by raising construction conditions. The market may be overpricing the policy shock, and technology stocks still have considerable room to rebound.
According to Goldman Sachs trading observations on November 5, some institutional investors reduced equity positions before the election to address uncertain risks, re-entered after the November 4 rally, but quickly sold when the market weakened the next day.
Goldman Sachs believes that the main selling pressure on Thursday came from macro trading products, not simply individual stock selling.
An intraday estimate from Nomura shows that leveraged ETFs may need to sell about $47 billion of stocks before the close for rebalancing, mainly involving technology and AI-related products.
Therefore, the stock price decline on November 5 may simultaneously include two parts: investors' reassessment of long-term AI returns, and the short-term trading shock formed by concentrated deleveraging.
04. What conditions are needed for the next reversal?
The stock market's adjustment to future investment growth may come earlier than changes in corporate financial statements. Even if some construction plans for 2028 are delayed, projects already under construction in 2027 may still continue to drive demand for chips, servers, and power equipment.
The short-term market is not without conditions for a rebound. Goldman Sachs observed near the close on November 5 that some investors were again focusing on Nvidia, TSMC, Microsoft, and CoreWeave. Stocks that had previously been sold off and amplified by concentrated deleveraging may recover quickly once funding pressure eases.
But a technical rebound is completely different from a true repair of the industry's investment logic. Such violent fluctuations show that the market has begun to doubt and reject the old logic.
The following indicators need to be closely watched next. First is AI revenue growth. Whether model companies, cloud service providers, and enterprise applications can maintain high growth rates and gradually improve operating cash flow will determine whether massive computing investment can obtain long-term returns. This is the source.
Second is data center delivery. Whether U.S. local governments truly tighten approvals and how much planned capacity can be completed on schedule will affect actual equipment procurement demand from 2027 to 2028.
At the end of its report, Citrini describes the current stage as an epilogue; the market's rhythm and tone are changing.
This may be a more accurate way to understand this crash.
One company's capital expenditure is another company's revenue. In the past, this relationship drove the entire industrial chain to rise together; today is different from the past, and the market is trading risk transmission along the same chain, falling together.
But in the future, trillions of dollars of investment will ultimately need an ending.
This article comes from the WeChat public account "Tencent Technology," author: Xiaojing, published by 36Kr with authorization.