November has been a brutal month for the AI trade. On November 5, two days after the US midterm elections, the Nasdaq fell more than 5% as investors worried that a new Democratic Congress and new state governments would move against AI and the data centers that power it. NPR put the losses across tech stocks at about $2 trillion in a single day. A week later, Nvidia’s chief executive, Jensen Huang, resigned after a personal scandal. That morning, Nvidia’s shares fell as much as 14%, briefly erasing up to $601 billion in value. Then on November 16, Gary Gensler, the former chair of the US Securities and Exchange Commission, told CNN that AI is “already a bubble” and will likely burst and cause a recession before the 2028 presidential election.
Every boom produces moments like this, and the question they raise is always the same. If the correction deepens into a crash, who gets hurt and who comes out stronger? The market has already started to answer. Over the past two weeks, investors have drawn a sharp line between the AI companies they still trust and the ones they are abandoning, and that line says a great deal about which parts of the AI economy are durable. It also has direct consequences for the companies that buy AI, from the price of the tools they use to the survival of the vendors they rely on.
The Nasdaq is 11.9% below its record. The dot-com crash went to 78%.
Before deciding who wins and who loses, it helps to be precise about how bad things are. On November 5, the Nasdaq-100 fell 6.4% in a single day. Chipmakers took the worst of it: AMD lost 17.3%, TSMC 10.8%, Broadcom 10.4% and Nvidia 9.3%, while Microsoft fell a more modest 3.6%. By the close that day, the Nasdaq Composite stood 11.9% below its October 6 record.
That is a correction, and a painful one, but it is a long way from a crash. When the dot-com bubble burst, the same index fell 78% over two and a half years. Nvidia itself has fallen further than the market. Its shares are down about 35% from their early October peak, roughly $2 trillion in market value, and at least seven Wall Street firms downgraded the stock in the week before Huang’s resignation.
Prediction markets have been pricing the risk of a full collapse. In late September, Polymarket traders put the chance of an “AI bubble burst” at 10% by the end of this year and 24% by mid-2027. The platform counts a burst only if at least three of six conditions occur within 90 days, including Nvidia closing 50% below its record high, the main semiconductor index falling 40% from its peak, and bankruptcies at major AI companies. Nvidia, at about 35% below its record, has not reached that threshold. The more useful question is what the sell-off has already revealed about where the strength and the weakness sit.
CoreWeave fell 15.4% in a week while Amazon rose 7.9%
The clearest signal came in the week of the election. Companies whose entire business depends on the AI build-out were hit hard. CoreWeave, the largest of the so-called neoclouds that rent out AI computing power, fell 15.4%, leaving it 56.6% lower than a year earlier. Nebius dropped 16.0% and Iren 9.7%. Among chipmakers, Micron lost 22.0% and Intel 14.4% in the same week.
The largest technology platforms barely flinched. Over the same week Amazon rose 7.9%, Alphabet gained 2.1% and Microsoft added 1.7%, while Meta slipped just 0.8%. Oracle was the exception among the big cloud providers, falling 11.2% for the week and 52.4% over the year. An analysis published in CounterPunch pointed to the reason: Oracle is the only one of these companies without a hugely profitable non-AI business to fall back on.
That split is the most important thing the sell-off has shown so far. The pattern suggests investors are separating AI companies that fund their spending from existing profits from those that fund it with borrowed money and future promises. Credit markets are telling the same story. One analysis of Bloomberg data showed the cost of insuring against default rising sharply for CoreWeave and Oracle, while it stayed calm for Amazon and Meta.
The likely losers borrowed to build. Short sellers hold 22% of one company’s shares.
If the correction becomes a crash, the companies most exposed are the ones that borrowed heavily to build AI infrastructure before the customers and revenue were fully in place. A Forbes analysis in late September, drawing on a warning from The Wall Street Journal, named the indebted neoclouds and data center builders, including CoreWeave, Oracle and SoftBank, as the likely losers. It estimated that in a burst CoreWeave could fall 40% to 60%, Oracle 25% to 35% and SoftBank 25% to 40%. The same analysis expected Nvidia and memory makers to suffer as well, though less badly where they carry little debt.
Short sellers, who profit when a share price falls, are already positioning for that outcome. They have focused on companies that combine heavy debt with uncertain timelines for cash flow. As of mid-November, 22.31% of Iren’s shares had been sold short, and the stock had lost nearly 53% of its value this year. Nebius had drawn 18.74% short interest after raising its 2026 spending forecast to roughly $27 billion and taking on a $3.5 billion secured financing facility from Citi.
Nvidia’s troubles add a second layer of risk for these companies. Reuters Breakingviews reported that before he left, Huang had built a web of deals across the industry. As of July, Nvidia counted $422 billion in future purchase, investment and lease commitments, plus up to $165 billion in credit and lease guarantees. Breakingviews called Nvidia the “central bank of AI” and warned that upstarts such as CoreWeave and Nscale face a rougher future without the executive who arranged much of that support.
The likely winners already make money elsewhere
Even people who believe AI is in a bubble tend to agree on who survives it. Gensler, despite his warning, told CNN that “the major companies will survive.” An investment newsletter from Stansberry Research, which also argues that a bubble has formed, made the same point more directly. It named Meta, Amazon, Alphabet and Microsoft as likely long-term winners because they are established businesses with large cash flows outside AI, and it described AI itself as a technology that will prove as transformative as the internet.
A downturn could make those companies stronger. When weaker competitors run short of money, the largest platforms can keep investing, hire talent the smaller firms let go, and buy promising technology at lower prices. Some suppliers outside the AI spotlight could also benefit. The Forbes analysis expected power equipment makers such as GE Vernova and cybersecurity firms to benefit from an AI bust.
The money has not disappeared either. GlobalSWF, a consultancy that tracks sovereign wealth funds, said it does not expect long-term investors to retreat from AI wholesale, and that lower valuations could let them increase their exposure selectively. Anthropic’s public offering, set to begin trading on November 19, is reported to be about six times oversubscribed even after the sell-off. Investors appear to be choosing more carefully, and they are still choosing AI.
Google is doubling the price of its newest AI models on December 1
For companies that buy AI, the most immediate effect of a tighter market may show up in their bills. Some of it is already scheduled. Google released Gemini 4 Carbon on November 10 at $2 per million input tokens and $10 per million output tokens. On December 1, the price of Carbon and its sibling Gemini 4 Argon doubles to $4 and $20. Two of Google’s Flash models follow on January 1, 2027, when their prices also double.
The pricing picture is mixed. Anthropic cut the price of one of its top models on November 9, and on October 26 OpenAI launched a new mid-range model at half the price of the model above it. The major labs have also converged on almost identical prices for their main workhorse models, at $2 and $10 per million tokens. Even so, Google’s move is a reminder that launch prices can change quickly once customers depend on a model.
A crash would put pricing to the test. Analysts have pointed to high valuations at AI companies that have yet to turn a profit, and those companies depend on investors who are willing to fund the gap. If that money becomes harder to raise, today’s prices could become difficult to sustain, and the tools that marketing teams have built into their daily work could cost far more to run. Signs of tighter funding have already appeared at the top of the industry. OpenAI has reportedly raised $11 billion in its latest round at a $1.25 trillion valuation, short of the $30 billion at $1.4 trillion it had sought, with the round still open.
Meta bought an AI startup, and its cloud service shuts down on November 20
A downturn also changes which AI vendors are still around a year from now. A wave of AI tools has launched over the past few years, many of them small companies built on top of models from OpenAI, Anthropic or Google and funded by venture capital. When money tightens, some of those companies run out of runway, some are bought by larger players, and some quietly shut down. Each outcome lands on the customers who built their workflows around the product.
A small example from this month shows how quickly that can happen. Meta bought Conductor, a startup whose software runs several AI coding agents in parallel. The team is joining Meta, and Conductor Cloud, the company’s hosted service, shuts down on November 20 and deletes the data stored on it. The deal says nothing about the health of the market, since acquisitions happen in good times too. What it shows is how little control a customer has once its vendor changes hands.
Vendors with thin finances have the least cushion when funding dries up, which is why the stability of an AI supplier deserves as much attention as its features.
The internet survived a 78% crash, and so did its strongest companies
The dot-com bust is the obvious comparison, and it holds a lesson about who wins. Between 2000 and 2002, the Nasdaq lost 78% of its value. Hundreds of internet companies disappeared, Pets.com among the most famous. Amazon’s shares lost more than 90% of their value between late 1999 and 2001, and the company survived to become one of the largest businesses in the world. The money and the valuations collapsed. The internet kept growing, and the companies with real customers and the means to keep going took over the market the others left behind.
The November sell-off already shows the same sorting at work. The companies that borrowed to build AI capacity ahead of demand are under the most pressure. The companies with large, profitable businesses beneath their AI spending are holding up, and investors are still lining up for the AI businesses they believe in. Two tests arrive this week. Nvidia reports its quarterly results on November 17, its first without its founder, and Anthropic begins trading on November 19.
Whatever those tests show, the technology itself is not going away. The question a crash would settle is which companies own it afterward. For anyone buying AI, that makes the strength of a vendor, the true cost of the tools and the stability of the platforms behind them as important as what the tools can do.
Sources
• CNN, Gary Gensler on the AI bubble, November 16, 2026
• The Weng Lab, "AI Bubble Burst Timeline," citing FRED and Yahoo Finance, November 2026
• Quantral, "Is the AI bubble bursting?" November 2026
• Forbes, "What Could Pop The AI Bubble And Which Stocks Stand To Lose," September 28, 2026
• The Washington Times, Polymarket burst conditions, November 12, 2026
• CounterPunch, "The Beginning of the End for the AI Bubble?" November 12, 2026
• Stansberry Research, "The Winners and Losers of the AI Bubble," October 2026
• Foreign Policy Journal, short sellers and AI data center stocks, November 16, 2026
• Reuters Breakingviews, "It's a bad time to unwind the Bank of Jensen Huang," November 16, 2026
• Semafor, sovereign wealth funds and the AI sell-off, November 10, 2026
• The Motley Fool, Anthropic IPO, November 16, 2026
• TokenCost, LLM API price changes November 2026
• AIonX, AI pricing comparison 2026
• Fello AI, AI pricing comparison, November 2026
• Compare AI Today, AI API pricing November 2026
• AI Business Weekly, AI Pricing Index, November 2026
• OneI AI Weekly, November 2 to 8, 2026, citing TechCrunch and SiliconANGLE
