The great Generative AI unraveling
Reported by Gary Marcus1 source ↓
Gary Marcus argues that falling valuations, rising costs and weak returns point to a downturn in generative AI. He says the economics of AI companies remain uncertain as competition increases.
Details
In his November 13 newsletter, Marcus says the economics of generative AI remain difficult: developing and running models is costly, while competition has pushed prices down. He argues that companies using similar techniques and data have little technical advantage over one another, making it harder to sustain profits.
Marcus cites reported declines over the past six months of 29% for Nvidia, 35% for Oracle and 54% for CoreWeave, and says Anthropic’s perceived valuation fell 25% in a matter of weeks. He also says the Claude 5.5 model family reportedly cost $10 billion to develop; the newsletter does not provide independent confirmation of these figures.
He links potential financial strain to Nvidia’s role in financing other companies, including cloud-computing providers, and speculates that a change in the company’s approach could leave some of them exposed. Marcus says it is unclear whether those businesses can remain viable without such support.
Written from coverage as of
Images
- Gary Marcus
- Gary Marcus
- Gary Marcus
- Gary Marcus
Stocks in this story
Change since open: 20 Nov to 21 Nov 2026
Last refresh