The AI trade broke on Friday, Nov. 20, 2026. The Nasdaq Composite fell more than 5% and turned negative for the year, while the S&P 500 entered correction, as Anthropic’s first day of trading as a public company failed to stop a market-wide sell-off in technology stocks, according to Yahoo! Finance.
Anthropic, which had defied the trend in its debut, fell back nearly to its IPO price during the session. Nvidia dropped double digits again. The chipmaker has lost about 10% of its value since Monday, a five-day slide that has pulled the entire AI complex with it.
What happened on Friday
The slump began in early October, right after the Nasdaq hit an all-time high. Since then, a Federal Reserve signaling higher rates, political fights over data centers after the midterms, and a widening gap between chip sales and chip stocks have fed the decline.
Chip sales are at record levels, almost 2.5 times what they were a year ago. The stocks are not. That disconnect is the core of the sell-off, and it points to a shift in bargaining power away from the companies that sell AI chips and toward the companies that buy them.
Google designs its own AI chips with Broadcom. On Friday, analyst Ming-Chi Kuo reported that Google cut about 20% of those chips from next year’s plans and moved some orders to a rival. Neither company confirmed the report. Broadcom shares fell more than 10%.
Kuo attributed the cut to electricity. Google does not have enough power for all the data centers it wants, so it is reserving capacity for the chips that run its best models. When you can only plug in so many chips, you can be picky about who you buy from.
Anthropic’s compute strategy under scrutiny
Anthropic is using its position as a buyer differently. Its models run on chips from several makers, so no single supplier has it locked in. Before the IPO, CEO Dario Amodei told investors he deliberately buys less compute upfront than he might need, then buys more at the last minute if demand runs hot.
That approach contrasts with OpenAI, which locked up large amounts of computing power early and used it to close much of the intelligence gap with Anthropic this year. Anthropic admitted in its IPO filing that it significantly underestimated how much computing power it needed, setting back its products.
Investor and Anthropic backer Mido attributed the strategy to the nature of AI research. Experiments are spiky. A lab runs small tests, and when one works it suddenly needs thousands more chips. Buying enough to cover every spike leaves many chips idle the rest of the time.
Mido called the current market a “GPU wastage bubble,” meaning many of the chips being bought are not getting used. If that is right, it is a problem for the companies selling them, and it helps explain why their stocks are falling even as sales hit records.
Anthropic still carries about $900 billion in long-term computing commitments, much of it going to the same chip makers. Two signals to watch: Nvidia’s margins, which should start to slip if buyers have the upper hand, and Anthropic’s costs, which should start to fall.
Private markets versus public markets
Carta CEO Henry Ward said the gap between private and public valuations has become enormous. Every AI startup is priced as if it will be the dominant winner. In public markets, only one or two actually are.
“In the private markets, you might have 50 startups doing the same thing. You’ve written them all as the winner,” Ward said. “49 of them will get written down because they won’t be the winner, and then the one winner will be written up.”
Ward said the reckoning for private companies comes when they go public and the entire capital structure converts to common stock, giving every investor in the world access to price it. He said Anthropic’s IPO raised the bar for everyone else.
“If you’re a $20 to $50 billion private software company, public markets are pretty tough right now,” he said. “You’re in the apocalypse trade.”
What the sell-off means for the AI trade
Dan Niles of Niles Investment Management compared the moment to 2000 to 2002, when the Nasdaq had seven rallies of 25% on the way to losing 78% of its value. He said the outcome is only knowable in hindsight.
Niles pointed to the midterms as the fundamental change. Forty of 41 ballot initiatives to halt data center construction passed, and governors in Texas, Georgia, Ohio, New Mexico and Michigan flipped on platforms to slow data center construction. If there is nowhere to put the chips, the chips stop selling.
He also flagged the economics of the two leading labs. Combined growth slowed from 75% in the second quarter to about 48% sequentially in the third, while operating losses rose from $7.9 billion to $8.4 billion as the labs compete on price against each other and against open-weight models.
“How are you going to fund $900 billion commitments if you’re Anthropic on $54 billion in revenues?” Niles said.
Data from Open Router shows the price of using models from Anthropic, OpenAI and Google down 57% since January, with usage up more than 13 times and spending up about 6 times. In the last five weeks, prices kept falling but spending flattened across both the big labs and cheaper open-weight models.
Factory co-founder and CTO Eno Reyes said customers are becoming more savvy about how tokens are spent, routing simpler tasks to cheaper models. He said the agentification of the economy has barely started, and that each person may eventually run thousands of agents.
The Nasdaq closed negative for the year. The question for Monday is whether Friday was a clearing event or the start of something longer. Niles covered his shorts into the close because his models said the market had reached maximum pain. He did not call it a bottom.
FAQ
Why did Anthropic’s IPO fail to lift the market?
Investors used the debut as an exit from the broader AI trade. Anthropic fell back near its IPO price, and the Nasdaq turned negative for the year despite record chip sales.
How much has Nvidia fallen?
Nvidia dropped double digits on Friday and has lost about 10% of its value since Monday, according to Yahoo! Finance.
What is the “GPU wastage bubble”?
It is investor Mido’s term for chips being bought faster than they are used. If correct, it pressures the chip makers’ margins and helps explain the sell-off in their stocks.
What happens next for the AI trade?
Watch Nvidia’s margins and Anthropic’s costs. If buyers keep the upper hand, the two should stop moving together. The midterm data center moratoriums also limit where new chips can be deployed.
