Midterm elections tear open the "abyss" of the AI trade, and the trillion-dollar capex narrative faces a moment of reckoning.

For months, investors have been worried about a series of risks behind the "cash cow" of artificial intelligence (AI) capital expenditure. This week, the market finally pressed the panic-selling button. After the U.S. midterm elections concluded on Tuesday, with Democrats winning majority seats in both the Senate and the House of Representatives while also scoring major victories in state legislatures and governors' seats, the market immediately began pricing in a worst-case scenario for the AI trade. Many winning candidates campaigned on restricting AI development, and the market fears that data center construction in many places may stall. This shift in expectations prompted traders to flee starting Thursday, with all assets tied to hundreds of billions in expected AI capital expenditure sold off. The Philadelphia Semiconductor Index plunged 12%, and the tech-heavy Nasdaq 100 fell 6.4%, with both indexes posting their largest one-day declines since the COVID-19 shock in March 2020. AI bellwether NVIDIA Corporation led the broader market lower, falling more than 9% and shedding more than US$440 billion in market value. Although reports later said that Texas Governor-elect Gina Hinojosa had made clear that all Texas data centers that have already received permits will be allowed to complete construction. Hinojosa plans to pass legislation requiring data centers that receive new permits in the future and are scheduled for delivery between 2028 and 2029 to pay taxes, commit not to drive up residential electricity prices, not affect residential water use, use union labor for construction, and give "Texans a seat at the table in decision-making." The news drove a premarket rebound in Nasdaq futures, but the market's doubts did not seem to dissipate. "The market is questioning the growth assumptions of data center supply chain companies, and there is a broad withdrawal of capital," said Peter Tuz, president of Chase Investment Counsel, whose fund holds shares in chipmakers NVIDIA Corporation (NVDA.US) and AMD (AMD.US), as well as major AI spenders such as Alphabet Inc. (GOOGL.US) and Amazon.com, Inc. (AMZN.US). "From the election results, it is easy to infer rising uncertainty in the regulatory environment, which in turn weighs on these companies' growth prospects." Since the start of the year, multiple pressures have continued to build, and market anxiety over AI capital expenditure has steadily accumulated: rising interest rates, growing debt levels for data center expansion, and increasingly strong public opposition to data center construction. AI capital expenditure has been the core engine of this round of gains in the S&P 500 and had once served as a buffer for the broader market against macro risks such as elevated oil prices, stubbornly high inflation, and the situation in Iran. Of course, given that the machinery of government often moves slowly, how efforts to curb data center development and regulate AI will evolve remains to be seen. But this round of selling is enough to show that the core pillar of this bull market is already mired in panic. "Right now the economy seems to be pulled by only one engine, and not only is that engine slowing, but the election results may slow it further," said Barry Knapp, managing partner at Ironsides Macroeconomics. "I wouldn't buy anything right now." According to estimates by strategists at UBS Group AG, nearly one-third of the computing capacity expected to come online in the United States by 2028 is now at risk. Most of that comes from data centers planned in Texas in the state's gubernatorial race, Hinojosa, who pledged to halt approvals for new data centers, defeated Republican incumbent Governor Greg Abbott to become the state's first Democratic governor since 1990. These concerns are prompting a reassessment of valuations in the already pressured AI sector. Take NVIDIA Corporation, for example, which has been the biggest beneficiary of AI investment. Its current price-to-earnings ratio is 13 times expected earnings over the next 12 months, down from more than 32 times a year earlier. Broadcom Inc. (AVGO.US), the second-largest U.S. chipmaker by market value, trades at 15 times forward earnings, below its peak of 42 times last December. "Previously, the market assumed that data center expansion was a long-term trend that could last for years," Knapp said. "If computing power construction can only last a few more years and the chip industry's long-term growth rate declines, then what valuation should we assign?" This round of selling has been especially brutal in the riskiest corners of the AI trade. AI computing services provider CoreWeave (CRWV.US) plunged 20% this week, down more than 75% from its peak in June 2025 and only a few dollars away from its March 2025 IPO price. Meanwhile, heavily indebted new cloud service provider Nebius (NBIS.US) fell 15% on Thursday, bringing its decline from its June high to 58%. Making matters worse, uncertainty over AI regulation and oversight of data center construction is making investors cautious about stepping in to buy the dip. "It's hard to judge whether the current decline has gone too far, because the regulatory outlook is still unsettled," said Luz of Chase Investment Counsel. "The market looks oversold, but we will wait for more clarity before considering buying." One benefit of slower data center construction is that tech giants such as Amazon.com, Inc., Alphabet, Microsoft Corporation (MSFT.US), and Meta (META.US) will be forced to cut spending. As recent post-earnings share price performance shows, investors can no longer tolerate capital expenditure continuing to exceed operating cash flow. Just on Wednesday, shares of these four giants plus Apple Inc. (AAPL.US) surged, adding more than US$800 billion in combined market value; but on Thursday, as the broader market fell, part of those gains were given back. "If they cut spending, maybe they won't have to borrow as much, and maybe cash flow will rise," Luz said. "But at the same time, growth may also slow. There are just too many questions right now."