On Wednesday morning, investors woke up to a decisive midterm result and did what markets usually do when uncertainty disappears: they bought. The Dow rallied 1,569 points, and the S&P 500 posted its best day in three months. But by Thursday afternoon, every bit of that optimism had been handed back, and then some, with a selloff underway.
The S&P 500 fell 4.4% on Thursday, its worst single day since the Liberation Day tariff shock of April 2025. The Dow lost 1,833 points. The Nasdaq Composite dropped 5.5% and closed in correction territory, nearly 12% below the record it set just last month. All eleven S&P sectors finished lower, and roughly $3 trillion in market value disappeared in a single session.
A day like that deserves a serious look. So let’s walk through what happened, why the damage landed where it did, what it could mean for the broader economy, and what history suggests about what comes next.
From Relief to Reckoning
The election itself was not ambiguous. Democrats took control of both the House and the Senate, flipping at least seven Senate seats in the process. Wednesday’s rally reflected relief that the outcome was clear and quickly accepted. Thursday reflected investors working through the implications, and one implication in particular: the AI infrastructure build-out (a.k.a. data centers).
Several of the most consequential races for markets weren’t in Washington at all. Governors-elect in Texas, Georgia, and here in Ohio campaigned on restricting data center development. And while I like to think Ohio is the heart of it all, Texas probably matters most in this case by a wide margin. Governor-elect Gina Hinojosa ran on pausing new data center projects until the state adopts tougher rules, and her party also won control of the Texas House. By one industry estimate, Texas alone was expected to supply close to 30% of the new U.S. data center capacity scheduled for 2028. Add a Congress likely to take a harder look at Big Tech, and the market reached a quick conclusion. The AI buildout may slow, get more expensive, and become more complicated.
And the backdrop wasn’t necessarily calm to begin with. The 10-year Treasury yield, while down from recent highs, still sits near 4.86%. Oil is hovering in the $90s, and futures markets are pricing in a 75% chance that the Federal Reserve raises rates in December. This morning’s jobs report (165,000 jobs added in October, far above expectations, with unemployment dipping to 4.1%) is great news for the economy. Still, it does little to cool those rate-hike expectations. When financial conditions are already tight, a policy shock lands harder.
Where the Damage Concentrated
So while the headline number was eye-popping on its own, the selling was not evenly distributed. The Philadelphia Semiconductor Index fell 12.3%, its worst day in roughly six and a half years. Nvidia dropped 9.3%, erasing its gains for the year and shedding about $447 billion in market value (in one day, from one company). AMD fell 17%, and Broadcom more than 10%. Combined, the 16 largest U.S.-listed chipmakers lost roughly $1.45 trillion.
The neoclouds fared even worse, relatively speaking. Remember, these are the companies that buy chips in enormous volume and rent out the computing capacity to AI developers and larger tech firms. CoreWeave fell roughly 10%, Nebius 15%, and IREN about 17%, after several of them had already lost as much as 40% over the prior month. CoreWeave is now down 76% from its June 2025 peak, its bonds trade around 81 cents on the dollar with a yield near 16%, and the cost to insure its debt against default hit a record this week.
The pressure extended well beyond chips. Tesla fell 11% and SpaceX roughly 9%. Caterpillar and Cisco, both meaningful suppliers to data center construction, each dropped about 6%. Selling carried overseas overnight as well, with Japan’s Nikkei falling nearly 6% into bear-market territory and Taiwan’s market down 8%.
Not All AI Exposure Is Equal
One of the more telling details from Thursday is what held up. The hyperscalers declined, but far less; Amazon and Meta, for example, fell in the 3% to 4% range.
The logic is straightforward. If new data center construction becomes harder, companies that already own permitted sites, signed power agreements, and operating facilities may find that existing capacity becomes more valuable, not less. The neoclouds, by contrast, depend heavily on borrowed money and on new capacity coming online on schedule, which leaves them most exposed to both higher financing costs and permitting delays. Industry analysts also point out that much of the 2027 buildout is already under construction. The real uncertainty sits in 2028 and beyond, where projects still need state approvals. As one strategist put it this week, local pushback is more likely to make the buildout pricier than to stop it outright.
That distinction matters. “AI” is not a single trade. It is a supply chain made up of companies with very different balance sheets, business models, and sensitivities, and Thursday priced those differences very quickly.
An Economy Leaning on One Pillar
There’s also a broader question here that extends beyond the stock market. AI infrastructure has become one of the most important engines of U.S. economic growth. Depending on how you count it, AI-related investment has accounted for anywhere from a third to three-quarters of U.S. GDP growth this year. Even at the low end, that is a remarkable concentration for a single category of spending. Third-quarter GDP grew at a healthy 3.4% annualized pace, with investment in information-processing equipment up more than 37%.
This is the part of Thursday’s selloff worth taking most seriously. If political resistance meaningfully slows data center construction, the impact won’t stay contained to semiconductor stocks. Fewer projects mean fewer orders for chips, servers, cooling systems, transformers, and construction labor, and a slower pace for the very investment that has been carrying a disproportionate share of economic growth. A pullback in AI spending, layered on top of elevated interest rates and expensive oil, is the kind of compounding effect that can turn a sector correction into a broader economic one.
To be clear, that is not the base case. The 2027 pipeline is largely locked in, the largest technology companies have repeatedly signaled multiyear spending commitments, and demand for computing capacity remains strong. But when a single engine is doing this much of the work, and policy change that makes it harder to run the engine is likely, it’s prudent to acknowledge the risk rather than assume it away.
What History Says
Now for the more encouraging side of the ledger. The well-known market research firm Bespoke Investment Group noted during Thursday’s selloff that both the Nasdaq 100 (tracked by the QQQ ETF) and the semiconductor sector (tracked by SMH) had fallen into extraordinarily rare territory: more than three standard deviations below their 50-day moving averages, with momentum readings firmly oversold. In plain terms, prices fell further and faster relative to their recent trend than they almost ever do. Since 1999, QQQ has only been more oversold on 0.2% of trading days.
Historically, these moments have been followed by strong returns. In the 16 prior instances, QQQ’s median return over the following year was about 18%, and it was lower a year later only twice. In the four instances where QQQ and SMH reached these levels at the same time, QQQ averaged a 25% gain over the next six months, and SMH averaged 68% over the following year.
Two caveats, though, are worth stating plainly. Four occurrences is a small sample. And one of the two exceptions was April 2000, at the start of the dot-com unwind, when oversold readings kept getting more oversold for the better part of two years. Extreme readings tell you that selling has been severe; they cannot tell you whether the underlying fundamentals have changed.
A Midterm Pattern
There’s another piece of history that applies here. As one well-known investor highlighted this week, in each of the 20 midterm elections since 1946, the S&P 500 was higher 12 months later, with an average gain of 14.8% compared to 8.7% for a typical year. The pattern held regardless of the outcome, including the nine times the president’s party lost control of the House, the Senate, or both.
That record includes some very difficult stretches. The 12 months following the 1986 midterms included Black Monday in October 1987, when the market fell more than 22% in a single day, and the S&P 500 still finished the period higher. Measured from the last close before Election Day, this year’s starting point was about 7,541 on the S&P 500. After Thursday, we sit roughly 2.5% below that line. That is a more uncomfortable start than usual, but the pattern has survived worse.
Discipline Over Reaction
Thursday was a severe day, and it exposed real vulnerabilities, particularly in the most leveraged and narrative-dependent corners of the AI trade. The political shift is real, the regulatory questions are real, and the economy’s reliance on AI investment is a genuine risk worth monitoring.
But severe days are also when investors are most likely to make decisions they later regret. Selling after a historic decline locks in losses based on the worst moment rather than the most likely path. The better response is the one that ideally was in place before Thursday: a diversified portfolio sized to your goals and time horizon, with no single theme, however compelling, carrying more weight than you could tolerate watching fall 10% in an afternoon.
Markets will keep debating how much this election changes the AI buildout, and that debate may take months to resolve. In the meantime, the most reliable path forward is the same one it has always been: a plan built to accommodate days like Thursday, and the discipline to stick with it.
Markets / Economy
- Markets thought they liked the election results, but a day later the AI-driven selloff more than offset the pickup. The S&P finished the week down -2.7%, the Nasdaq down -3.4%, and the small-cap Russell 2000 down -2.1%.
- The U.S. economy added 165K jobs in October, following an upwardly revised 60K increase in September, and well above market forecasts of 65K. The figure is also higher than the average monthly gain of 48K over the prior 12 months.
- Unemployment in the U.S. eased to 4.1% in October, down from 4.2% in September and slightly below market expectations of 4.2%.
- U.S. job openings unexpectedly rose by 333K to 7.523 million in September, from an upwardly revised 7.19 million in August, beating expectations of 7.10 million and reaching a four-month high.
Stocks
- U.S. equities were in negative territory. Energy and Technology led the decline, while Communication Services and Real Estate outperformed. Value stocks led growth stocks, and small caps beat large caps.
- International equities closed lower for the week. Developed markets fared better than emerging markets.
Bonds
- The 10-year Treasury bond yield decreased eight basis points to 4.86% during the week.
- Global bond markets were in positive territory this week.
- Corporate bonds led for the week, followed by government bonds and high-yield bonds.
