Tech • AI • Robotics • Game

ENFR

It Happened... The AI Bubble Just Cracked (Worst Week for AI Ever)

9.4/10
AIAI RevolutionNovember 22, 2026 at 01:24 AM14:00
Audio player

⏱ ≈ 4 min read · 14-min video — 10 min saved

TL;DR

Anthropic’s blockbuster market debut, a historic selloff led by Nvidia, fresh scrutiny of AI supply chains and a claimed AI-assisted breakthrough on the Riemann Hypothesis underscored how rapidly enthusiasm and anxiety around artificial intelligence are colliding.

KEY POINTS

Anthropic’s huge IPO

Anthropic sold 980 million shares at $83 and raised $81.3 billion in a record IPO, then added shares the next day to lift the total to $93.5 billion. The stock opened at $111.50, valuing the company near $2 trillion, before closing around $97.17 for a market value of about $1.8 trillion, still above Meta and Tesla. A separate $26 billion private placement included $10 billion each from Nvidia and Google, with additional stakes taken by AMD, Broadcom, Micron, Microsoft and TSMC.

Cross-ownership fuels bubble concerns

The deal highlighted a tight feedback loop in AI financing: the companies supplying chips and cloud infrastructure are also becoming equity holders in the labs buying their products. Anthropic has pledged more than $1 trillion in cumulative computing spend, intensifying debate over whether AI valuations are being sustained by circular capital flows rather than proven revenue. The concern is no longer only how much companies can spend on AI, but whether those investments can produce returns commensurate with the scale.

Protests over recursive self-improvement

More than 100 protesters gathered outside Anthropic’s San Francisco offices, chanting “Ban AI that builds AI.” Demonstrators linked to the Ban RSI campaign warned against recursive self-improvement, the prospect of AI systems helping design more capable successors at a pace humans cannot effectively monitor. Similar protests in New York led to arrests outside the Nasdaq building.

A weak second day and broader market shock

On November 20, Anthropic shares fell about 8% to 12%, at one point trading only slightly above the $83 IPO price. The decline came amid the sharpest AI stock rout in years rather than any immediate company-specific miss. The drop echoed a broader investor reassessment of whether the sector’s growth narrative can justify its valuations.

Nvidia posts a record weekly loss

Nvidia, the central hardware supplier to the AI boom, fell 14% on November 20, wiping out about $485 billion in a day. Over the week, it lost nearly $900 billion, the largest one-week value destruction ever recorded by a company. The stock has now fallen for seven straight weeks, down 46% over that stretch, erasing more than $2.5 trillion in market value.

Regulatory and governance pressure at Nvidia

The selloff accelerated after reports that the SEC was examining potential accounting issues. Questions have centered on revenue routed through Singapore, where Financial Times reported Nvidia booked about $104 billion in 2025 sales, and on alleged links between a Singapore company called Mega Speed and chip flows into China. California’s attorney general also subpoenaed Nvidia over whether chips or trade secrets reached China and whether investors were misled.

The slump spread across global markets

AI-linked names across Asia and the US fell sharply. SoftBank dropped about 14.5% in Tokyo, Samsung fell 7%, SK Hynix nearly 6%, and TSMC almost 5%. In the US, the S&P 500 lost 3.7%, the Nasdaq 100 fell 5.6% in a day and more than 10% for the week, its worst performance since March 2020.

US-China policy signals remain mixed

Treasury Secretary Scott Bessent said Donald Trump and Xi Jinping agreed that humans should remain involved in any military decisions made by advanced AI and that no major military operation should be launched without human involvement. No written framework was released. At the same time, Trump said the US could heavily tax companies building data centers abroad, even as large American tech groups already operate them internationally.

Congress documents China chip approvals

The House Foreign Affairs Committee released documents showing the Commerce Department approved 879,000 Nvidia H200 and AMD chips for Chinese companies between December and May, out of 3.4 million requested. The approvals included 150,000 for Tencent, 75,000 for Alibaba and 43,000 for Baidu, all named on the Pentagon’s list of alleged Chinese military companies. Nvidia said it has since stopped shipping H200s to China.

AI research claims a major mathematical breakthrough

On the same week markets convulsed, Levent Alpoge, a mathematician at Anthropic, said work with Claude helped produce a proof of the Riemann Hypothesis, a 167-year-old problem and one of the Millennium Prize questions worth $1 million. The proof was formalized in Lean, with a human-readable version reportedly exceeding 250,000 pages. MIT mathematician Andrew Sutherland said the chance of something being fundamentally wrong appeared close to zero, though full verification was still ongoing.

CONCLUSION

The past two days captured the central tension of the AI era: extraordinary technical claims are arriving just as investors and regulators become far less willing to accept extraordinary valuations on faith.

Ask a question

More from AI