The Thursday the market read the election twice: the AI data-centre sell-off
On Wednesday 4 November, the day after the US midterm elections, the stock market rallied. The Nasdaq Composite rose 1.4% to 25,736.51, as if the result had settled something. On Thursday, the market read the same result again — and the AI data-centre sell-off began.
The Nasdaq Composite fell 5.52% to 24,315.00, its worst single day in 19 months and its deepest drop since April 2025. The S&P 500 fell 4.36% to 7,353.73, wiping roughly $3 trillion from the value of its companies, according to Reuters. The Dow dropped 3.44% to 51,523.06. The Nasdaq 100 fell about 6.36%, losing approximately $2.5 trillion in market value. What broke was the price investors were willing to pay for the boom.
The selling sat exactly where the boom lives. The Philadelphia Semiconductor Index fell 12.25% to 9,011.14, its second-worst fall on record. Nvidia fell 9.26% to $181.60, erasing its gains for the year. AMD fell 17.26% to $404.40. Micron fell about 17%, SanDisk 22%, Broadcom 10.43%, Qualcomm 9.5%. The fear, as the market priced it, was not about chips or demand. It was about concrete, steel and electricity — and whether the politics of building data centres had just changed.
What the voters actually did

On 3 November, Democrats took control of both chambers of Congress. They also flipped governorships in three states. In Texas, Gina Hinojosa defeated Governor Greg Abbott after campaigning for an immediate moratorium on new data-centre permits. In Georgia, Keisha Lance Bottoms won after promising similar restrictions. In Ohio, Amy Acton won on a pledge to pause development.
It was not only the statewide races. Ballot measures limiting data centres passed in 40 of 41 contests across the country, according to an Al Jazeera count. An NBC exit poll found 56% of voters called AI and data centres a crucial or major factor in their vote. The numbers behind the politics explain the market’s alarm: Wells Fargo puts Texas at 27% of planned US data-centre gigawatts, and Goldman Sachs estimates that Democratic statehouse wins in Texas, Michigan and Georgia could put roughly 45% of planned US data-centre capacity at risk from construction moratoria, tighter regulation and reduced tax incentives.
None of the campaign pledges is law yet. The Texas legislative session that would decide a state moratorium does not open until 12 January. Senator Bernie Sanders has called for a national moratorium, but stopping the buildout remains, in J.P. Morgan’s phrase, “a tall order”. What moved trillions of dollars last Thursday was a promise — or rather, several promises, made on a campaign trail, about policies that do not yet exist.
The calm voice in the room
That is the point Aneri Jambusaria, group managing director and chief wealth officer at LPL Financial, made in conversation with the Economic Times. Her counsel to investors, as the paper reported it, was to wait: wait for political promises about curbing data-centre investment in the United States to translate into actual policy before drawing conclusions about the AI trade. LPL’s investment leadership, she indicated, is not yet convinced the AI rally is in bubble territory.
It is a deliberately unexciting position, and that is what makes it interesting. J.P. Morgan’s analysts put the market’s behaviour in a single sentence: “Before the policy details have even been settled, or formally proposed, investors have already repriced these possibilities.” The market did not wait for the policy. It priced the promise in one session — and then, on Friday, began to take some of it back. The Nasdaq gained 1.09% to 24,579.42, the semiconductor index rose 4.21%, and AMD rebounded 11.04% to $449.04. The week still ended down, but some of Thursday’s fall was taken back on Friday.
Or maybe the market was looking for an excuse

There is a second reading of Thursday, and honesty demands it. A 5.52% fall in the Nasdaq after a long rally can be ordinary profit-taking and valuation nerves, with the midterm result serving as a convenient trigger rather than the cause. The evidence for that reading is not hard to find.
Start with demand. Two days before the sell-off, AMD reported third-quarter revenue of $13.38 billion, with data-centre revenue up 97% year on year to $8.55 billion, and guided fourth-quarter revenue to about $17 billion. Whatever the politics, customers are still buying the hardware. Then look at valuations: J.P. Morgan notes that the 24-month forward price-to-earnings ratio of the MVIS US Listed Semiconductor 25 Index has halved since June, from 22.5 times to 11.4 times. And the analysts who cover the sector for a living were notably calmer than the tape: Morgan Stanley and UBS said on 9 November that the sell-off was overblown, that data-centre regulatory risks remain low, and that policy changes were likelier to delay some projects from 2028 to 2029 than to halt capital spending.
If that reading is right, then “wait for policy” is sensible advice about a market that was looking for an excuse to fall — and found one in the election. The two readings are not really in conflict. The sell-off can be both a genuine repricing of political risk and a bout of profit-taking wearing a political costume. Either way, the thing being sold was a story about the future, not a fact about the present.
What would actually change the story
Here is what to watch, now that the shouting is over. The first test is Texas, where the new governor’s moratorium promise meets a legislature that convenes in January — and where the form of any moratorium remains an open question: a conditional pause, a construction ban, or new cost, water or power conditions on permits. The second test is whether the pauses Goldman Sachs expects to be short-lived stay short-lived, or harden into something durable. The third is whether the companies that already own facilities — the hyperscalers J.P. Morgan says have the edge when regulation tightens — start to benefit from the scarcity that scares their competitors.
Nigel Green, chief executive of deVere Group, captured the shift in a line that will probably outlast the week: “For two years, the AI trade has been valued on chips, software and demand. From this week, it may also have to be valued on permits, power tariffs, borrowing costs and political risk.” He may be right.
That is the wry heart of the episode. Trillions of dollars moved on a promise, in a single session, before any of those pledges had become law. The calm voice in the room — the one saying wait and see what actually becomes law — was not the exciting voice. Whether those pledges become policy, in Texas or in Washington, remains to be seen. Until they do, the market has priced a story it has not yet read.












