The selloff was about capacity, not just prices
On Thursday 5 November the Nasdaq Composite was down 5.3% intraday at the 1:01 PM EST update, in correction and more than 10% below its record from last month. The S&P 500 was down about 4%, its worst day since April 2025, and the Dow fell 1,657 points, or 3.1%. Those are live-blog readings, and they differ between outlets. StockTitan recorded the Nasdaq at -3.9% at 11:49 a.m. ET, so treat any single figure as a snapshot, not a close.
The damage was concentrated where builders spend money. The iShares Semiconductor ETF (SOXX) fell close to 11%, its worst day since March 2020 and down 15% for the week. AMD and Micron were each down about 15%, CoreWeave about 9% and Nebius about 13%. Nvidia fell about 6% to $188.75.
The trigger was political, not earnings. Democrats swept the midterms and will control both chambers of Congress from January. Investors feared the sweep would slow the data-center buildout, and OpenAI reaffirmed its pledge to pay the full cost of the power its data centers use. The Motley Fool reports that the Texas governor-elect had campaigned on pausing new data-center construction, and that Dell fell about 9% and Applied Digital about 17%.
The useful read is not that chip stocks are volatile. Builders have been assuming that GPU and cloud capacity keeps arriving on schedule and at stable prices. A stock move is the market's opinion of that assumption. A delayed data center is a planning input, and it lands on your roadmap, not just on a trading screen.
Local opposition is now on the ballot
The backlash is not only in Washington. Fortune reports that Ohio voters decided 21 data-center ballot measures in 18 municipalities on 3 November. That is more than half of the 41 such measures nationwide, and at least 14 of those municipalities passed bans or other restrictions.
Public opinion points the same way. Fortune cites a Gallup poll from May finding that about 70% of Americans opposed data centers in their local area. It also cites an ICAPP/SurveyUSA poll of 1,650 Ohio adults, in which 55% opposed data-center construction and 22% supported it.
These are opinion measures and one state's ballot results, so they do not prove that any particular project will be stopped. Still, local zoning and ballot outcomes are where a build-out gets slowed in practice. A data center needs land, power approvals and permits long before it needs a single GPU, and each of those steps can be contested by the people who live nearby.
The number to plan around is a forecast
The figure planners should look at first comes from UBS. Its strategists estimate that about 30% of data-center additions expected in 2028 are now at risk. I could only see that estimate through StockTitan's summary of a Reuters report, so treat it as a secondhand forecast, not a measured shortfall. StockTitan's live-market write-up is where I checked it.
Politics is moving in the same direction. Senator Bernie Sanders has called for a national data-center moratorium and wrote on X that "Americans don't want Big Tech oligarchs wrecking the environment, gutting jobs and threatening humanity's future." A call for a moratorium is not a law. Its significance is that it sets the agenda for a Congress that will be controlled by the party that just won.
Investors are watching the same thing. Tim Holland, chief investment officer at Orion, said investors are "stepping away until they get a better sense of the implications of the midterms around the AI capex cycle in particular." Keith Lerner, chief investment officer at Truist Wealth, said: "The rubber band is becoming increasingly stretched."
What is still unknown
Most of what matters here is not settled. We do not know how much of the 2028 capacity will actually slip, or whether the incoming Congress will pass anything that restricts construction. We do not know whether local bans will spread beyond the Ohio ballot measures, or whether OpenAI's pledge to pay the full cost of its power becomes a pattern across the industry.
The intraday market numbers will change, and the forecasts may be revised in either direction. What the week does establish is that the market now prices political risk into AI infrastructure. Whether that risk is large or small, it is no longer hypothetical.
Audit where your capacity assumptions live
Treat this as a planning exercise, not a prediction. Five steps are enough to find your exposure.
1. Inventory every workload. For production inference, batch evaluation, fine-tuning and training, write down the region, the provider and the commitment behind each one. Reserved capacity, committed spend and on-demand usage are different risks. Mark which workloads break if capacity arrives six months late, and which only get more expensive.
2. Read the contracts for the day something slips. Find the notice periods, the minimum commitments and what the provider owes you when a region is constrained. If nobody on your team can answer those questions from memory, you have found a risk worth an afternoon.
3. Make routing portable. Put model and provider choice behind one internal interface, selected by configuration rather than code. Keep at least one fallback model that you exercise in staging on every release. A fallback that exists only in a runbook has never been tested. Keep its eval results current so a switch can be made on evidence.
4. Track cost per task, not only cost per token. If capacity tightens, prices move, and a cheaper model you cannot reach is worth less than a dearer one you can. Put budget alerts on the workloads from step one, and review them against the outcomes they produce.
5. Decide your triggers now. Write down what would make you move a workload before the news arrives: a provider delaying a region you depend on, a cost increase above a threshold you choose, or a local restriction near a site you rely on. A trigger written in advance means you act before the release is blocked, not after.
The selloff will be read mostly as a market story. For builders it is a reminder that the capacity you plan on is a bet on politics as well as engineering. The bet is cheaper to change this quarter than it will be after a delay lands on your release calendar.
Comments (0)
No comments yet. Start the conversation.