Gary Marcus: The Generative AI Bubble Is Popping as Nvidia Falls 29% in Six Months

Gary Marcus · rss · 2026-11-14

Gary Marcus argues the generative AI financial bubble is finally near its end. He revisits his warnings since August 2023: everyone followed the same techniques and similar training data, so no one has a real moat, price wars are inevitable, and profits stay elusive.

Key arguments

Training and inference are extremely expensive; the Claude 5.5 family alone reportedly cost $10B to develop, leaving little room for profit.
LLMs remain unreliable, with studies showing weak ROI; a massive Microsoft outage is suspected to stem from AI-written code, and the Klarna-style rehiring of humans is another sign of overhype.
He concedes he was wrong about timing: hypey valuations lasted far longer than expected.

Market signals

Nvidia is down 29% over six months, Oracle 35%, CoreWeave 54%, and Anthropic's perceived valuation fell 25% in weeks.
Mid-CNBC hit, a breaking bulletin announced neocloud N-Scale was in deep trouble after overcommitting.
He suggests a leadership change at Nvidia may end its role as everyone's banker; if it stops propping up neoclouds, several firms are in trouble and Anthropic's IPO could be jeopardized.

Bottom line: OpenAI, Anthropic, Nscale and CoreWeave have never been demonstrably profitable, and now may have to stand on their own.

Original post →

Gary Marcus: The Generative AI Bubble Is Popping as Nvidia Falls 29% in Six Months

More from AGI Musings