In brief

• Nvidia gained 2.78% and AMD jumped 11.04% Friday, recovering only part of Thursday’s brutal losses.
• The turmoil spread to Asian chipmakers, prompting efforts to stabilize Taiwan’s market.
• Data-center restrictions and the huge cost of AI infrastructure are testing investors’ confidence.

AI stocks staged a comeback Friday after a rout that swept from Wall Street to Asia. But a single green session hardly settles the bigger question for investors heading into the weekend: Is this an overdue correction, or the beginning of an AI bubble bursting?

Nvidia rose 2.78% to $186.64, after plunging 9.3% this Thursday. AMD also bounced to $449.04 following a 17.3% collapse a day earlier. Both stocks remain well below their October peaks, showing how little a sharp relief rally can repair.

It wasn’t just U.S. markets seeing red either. South Korea’s KOSPI lost 6% Friday, while Taiwan’s benchmark dropped more than 8% during the week—its worst week in months. TSMC hit its daily limit-down threshold during the rout. Reuters reported that Taiwanese officials asked state-owned banks to help support prices.

The US Tech 100 cash-market proxy also recovered, gaining 1.85% to 27,007.3, which means it’s down more than 5.5% this week. Its daily candles show the bounce arriving after a much larger slide from October levels, not a return to the previous trend.

Why the AI trade is unraveling

The U.S. midterms accelerated fears that Democratic gains would mean tougher scrutiny of data centers. In 40 of 41 local ballot contests on November 3, voters favored limiting development, including by rejecting pro-construction measures, Ballotpedia found. Some votes were advisory, but the political message was unmistakable.

That matters for chip companies because data centers are their biggest growth story. Each project requires vast spending on processors, servers, power and cooling. Restrictions could delay orders or force construction elsewhere.

Yet the doubts predated the election. AI companies and their customers are spending aggressively to build capacity, often before returns are clear. More expensive borrowing makes that bet harder to sustain. Reuters quoted market strategist Steve Sosnick warning that heavily owned stocks were already “priced to perfection.”

Correction or bubble burst?

A correction means a market has fallen at least 10% from a recent peak. It does not necessarily mean the underlying business is shrinking. Imagine a chipmaker whose shares trade at 60 times annual earnings because investors expect explosive growth. If investors become willing to pay only 40 times those same earnings, the stock can lose a third of its value without profits falling at all.

A bubble bursting is more serious: It happens when expectations prove unrealistic and the money supporting the boom begins to disappear.

History offers a useful warning. During the dot-com crash, the Nasdaq lost roughly 77% between March 2000 and October 2002. The internet was not a fad, but investors had vastly overestimated how quickly companies could turn it into profit.

Telecom operators spent fortunes laying fiber-optic cables for customers who had yet to arrive. Global Crossing, which built an international network, filed for bankruptcy in 2002. The technology survived; many companies and shareholders did not.

AI could face a similar mismatch between long-term promise and near-term returns—and some investors are already talking about it. For example, the RAM shortage is also a business opportunity that many companies are trying to use for their advantage. Former crypto miners are renting their hardware and cloud provider and inference startups are appearing everywhere.

Now, consider a hypothetical company borrowing billions to build a data center packed with Nvidia chips, expecting to rent that computing power at premium prices for years. If customers spend less than anticipated, borrowing becomes more expensive or construction is delayed, the debt still has to be repaid.

The operator may cancel chip orders, its contractors lose work, and utilities that planned new capacity lose anticipated customers. Lenders and investors holding its debt could also take losses. A stock-market selloff can then become a problem for the wider economy.

There is an important difference from 2000, however. Microsoft, Alphabet, Amazon and Meta are profitable businesses with established customers and substantial cash flows, unlike many of the speculative dot-com startups. They can withstand setbacks better than smaller AI-cloud providers that depend on repeatedly borrowing to expand.

It is possible for AI to keep transforming industries while a bunch of individual AI companies fail—and for investors in successful companies to have overpaid.

The financial warning lights are beginning to flicker. Reuters reported Friday that the cost of insuring against a default by AI-cloud provider CoreWeave had climbed above 1,000 basis points. Such protection works somewhat like insurance on a loan: as the price rises, investors are signaling greater concern about repayment.

Yet broader bond markets were still operating relatively normally. That points to growing strain in the riskiest corners of AI financing, not proof of a system-wide crisis.

Next week will offer more concrete tests than another day of red or green candles. Results from CoreWeave, Applied Materials, and Cisco could reveal whether customers are still ordering capacity and equipment, while inflation data will influence expectations for borrowing costs. Stable orders and manageable debt would support the correction argument.

Repeated canceled projects, failed refinancing and shrinking sales would suggest the boom's financial foundations are cracking. The question is not whether AI has a future, but whether every company betting billions on that future can afford to get there.

Daily Debrief Newsletter

Start every day with the top news stories right now, plus original features, a podcast, videos and more.