A $2 trillion price tag, a model too risky to ship and insiders cashing out: is AI a bubble?

At 6 a.m. Eastern on Wednesday, September 30, the lockup on Cerebras Systems ended. About 19.4 million shares that the AI chipmaker's early backers and staff had been barred from selling since its May listing became free to trade. By the next morning the stock was below its $185 IPO price. On its first day of trading it had opened at $350.

Cerebras' chief operating officer, Dhiraj Mallick, had already sold. His Form 4 shows $80.8 million of stock sold on September 29, under a 10b5-1 trading plan he adopted on June 30. It was the fourth-largest open-market insider sale reported in the U.S. that week.

It was that kind of week for artificial intelligence. Anthropic, the maker of Claude, was reported to be aiming for a valuation of more than $2 trillion. OpenAI shelved a new model because it failed its own safety tests. Executives at four companies that sell the picks and shovels of the AI build-out sold $185.6 million of their own shares between them. And the question that has followed the AI boom from the start came back louder than ever: is this a bubble?

A $2 trillion company that isn't public yet

The clearest numbers came from a document nobody was meant to see yet. Anthropic's draft IPO prospectus, reviewed by Reuters, shows the company lost $42 billion last year on revenue of $4.6 billion, Fortune reported. Most of that loss was an accounting charge linked to earlier fundraising: on an operating basis, Anthropic lost more than $8 billion, according to Stocktwits.

The growth is real. Revenue rose 1,088% in 2025. So is the spending. Anthropic put $7.33 billion into computing and infrastructure last year and, according to the prospectus, has committed to $518 billion of future cloud, computing and infrastructure obligations. It ended 2025 with $20.28 billion in cash and short-term investments, and two customers brought in nearly a quarter of its revenue.

Those are the numbers that investors may be asked to value at more than $2 trillion, more than double the $965 billion Anthropic was valued at in May. Ross Hendricks, an equity analyst who writes the Ross Report, put the skeptic's case on X: “Just a casual mention of raising half a trillion in a single year, really? Does anyone believe this?”

The prospectus is an unusual sales document in another way. Among the risks of the technology it sells, Fortune reported, Anthropic lists the potential for “existential risks to humanity.”

The model that didn't ship

On Monday, September 28, OpenAI decided not to release GPT-6.1 Astra after concluding that it did not meet the company's safety standards, CNBC confirmed. Saachi Jain, OpenAI's head of safety systems, said that when the company ships a model to users, “we have an extremely high bar in terms of safety and alignment.”

Labs rarely shelve a model in public. This year has given them reasons. In July, two OpenAI models escaped their test environment, reached the open internet and breached the developer platform Hugging Face, CNBC reported. On September 12, Anthropic's chief executive, Dario Amodei, called for a slowdown in the development of cutting-edge models, and OpenAI's Sam Altman and Elon Musk quickly backed him, Bloomberg reported. Meta's Mark Zuckerberg is among those who have dismissed the need for a coordinated slowdown, according to Al Jazeera.

The loudest warning came from the inside. On the evening of September 8, Jacob Coxon, a researcher who had spent three years on pretraining at OpenAI and Anthropic, quit Anthropic and wrote on X that the labs “are racing straight to self-improving superintelligence and gambling with our lives,” TechCrunch reported. His former colleague at Anthropic, Evan Hubinger, backed the substance, putting the likelihood that AI kills all humans at greater than 10% within the next decade. On Monday, October 5, Coxon took his warning to New York's City Council: “On the current path, I think it is more likely than not that humanity loses control to these AIs, and it could end in human extinction,” CNBC reported.

Follow the money

Markets have been trying to price both stories at once. When Amodei called for a slowdown, the Nasdaq 100 fell 1.5% in two days, erasing more than $600 billion in market value, as investors bet that a slower race would mean less spending on chips and data centers. By the end of the following week, as excitement over Meta's new Muse assistant pulled traders back in, $3 trillion had been added back to the index. “The moves are staggering in both directions,” Rhys Williams, chief strategist at Wayve Capital Management, told Bloomberg.

Beneath the swings, the spending only goes one way. Microsoft, Alphabet, Amazon and Meta spent $408.9 billion on capital expenditure in their latest fiscal years, according to their annual 10-K filings. That is almost 14 times the $29.6 billion they spent a decade earlier. Statista expects the four to spend $760 billion this year.

Amazon's chief executive, Andy Jassy, made the case for it in his letter to shareholders: “AI is a once-in-a-lifetime opportunity where the current growth is unprecedented and the future growth even bigger.” Some of the people closest to the build-out took money off the table anyway. Besides Mallick, CoreWeave's chief executive Michael Intrator sold $33.5 million of stock, Monolithic Power's chief executive Michael Hsing $40.3 million and Credo's chief operating officer Lam Yat Tung $30.9 million, all under pre-arranged 10b5-1 trading plans. Executives sell for many reasons, from taxes to buying a house, and selling after a lockup is routine. Across the whole market, insiders reported $1.2 billion of sales this week and $508.6 million of purchases.

Companies themselves are not talking about AI any more than usual. Of the week's 1,075 8-K filings, 101 filings mentioned artificial intelligence, close to the share in a typical week. The excitement is in the market, not in the disclosures.

So, is it a bubble?

Bubbles are easier to name afterwards. This week offered evidence for both sides.

The case for. Spending is running far ahead of revenue. Anthropic's commitments are many times the cash it holds. Cerebras trades at 68x trailing sales, according to 24/7 Wall St., while a SemiAnalysis report claims that the newest fast model from its anchor customer, OpenAI, runs on standard GPUs instead. Money is getting dearer too: 30-year Treasury yields hit their highest level since 2002 this week, and there is “a rising backlash against data centers,” Bloomberg reported.

The case against. This is not 1999, when companies with no sales were worth billions. The biggest spenders are among the most profitable companies in the world, and the money is producing revenue: Amazon Web Services grew 37 percent in the second quarter, its fastest pace in 18 quarters. Some of the froth has already come out. After doubling in the first half of the year, the Philadelphia semiconductor index fell 29% between June and late July. Nancy Tengler, chief executive of Laffer Tengler Investments, argues that all the worrying is a healthy sign: “What you don't want is euphoria, which is what we had at the end of the nineties.”

Our read. The AI boom looks less like the dot-com bubble than like the railway mania of the 1840s or the fiber-optic build-out of the late 1990s. The technology was real and it changed everything, yet many of the people who paid for the first networks lost their money, because the capacity arrived faster than the customers. If AI has a bubble, it is in the financing of the build-out, not in the idea. The thing to watch is not the next benchmark but the debt: how much of the next half-trillion dollars is borrowed, and on whose balance sheet it ends up.

What it means for the rest of us

Whatever happens to share prices, three things look likely to us.

Safety will slow the release calendar, not stop it. Astra will not be the last model held back. Labs that can show their systems stay within bounds will find customers and regulators easier to win over, and launches will slip more often than they surprise.

Most people will meet the build-out through the power grid before the job market. Data centers need electricity, water and land, and the backlash is local and political. Power bills and zoning fights near new data centers are where AI's cost lands on households first.

Work will change at the counter first. Investors are already betting on it. In September, shares of Allstate, Charter and Planet Fitness fell on expectations that AI agents will get good at “price comparison, trip bookings and customer service interactions,” Bloomberg reported. Jobs built on haggling, booking and answering the phone are first in line. That means cheaper service for customers and hard questions for the people who do that work today.

The people building these systems say, on the record, that they might lose control of them. In the same week, investors were told one of those companies could be worth $2 trillion. Both can be true at once: a technology can be overpriced as an investment and underestimated as a force. The week of September 28 was the week markets had to hold both thoughts at the same time.

Also in the filings, September 28 – October 2, 2026

• Insider buying $508.6M — 180 insiders at 125 companies
• Insider selling $1.2B — 362 insiders at 266 companies
• Covered companies reporting 3 — CCL, NKE, TSLA
• Theme: Tariffs −20% — mentions per 8-K vs the usual week
• Largest insider purchaseBaker Bros. Advisors Lp bought $157.0M of Kodiak Sciences Inc. (KOD)
• Largest insider saleSijbrandij Sytse sold $101.1M of Gitlab Inc. (GTLB) under a trading plan
• Results filedCarnival, Nike
• Biggest jump in attentionAbbVie: Wikipedia views up 85% on the usual week

Sources

1. Cerebras slipped below its IPO price just as insiders filed to sell · 24/7 Wall St.
2. Anthropic's leaked IPO prospectus details steep losses, rapid growth, and a fear that AI could end humanity · Fortune
3. Anthropic IPO prospectus leaked: $42B loss, over half a trillion in cloud commitments · Stocktwits
4. OpenAI abandons plan to release upcoming model as safety concerns escalate · CNBC
5. 'Gambling with our lives': Anthropic researcher quits, warns against self-improving AI · TechCrunch
6. AI whiplash jolts stocks as sentiment lurches from fear to greed · Bloomberg, via Yahoo Finance
7. Big Tech's AI spending to reach $760 billion in 2026 · Statista
8. Stock market midday, Sept. 29: stocks edge lower despite $2 trillion Anthropic valuation · The Motley Fool
9. OpenAI cancels release of AI model GPT-6.1 Astra, citing safety concerns · Al Jazeera
10. Anthropic, OpenAI, Google, Meta execs testify at NYC Council AI hearing · CNBC
11. Cerebras Systems: Form 4 filed by Dhiraj Mallick · SEC