Thom Wolf on AI Compute Centralization and Oligopoly Risks

stanfordnlp · x · 2026-08-28

Thom Wolf highlights the final 10 minutes of the Dwarkesh podcast, which discusses the trend toward extreme concentration of power in AI. OpenAI and Anthropic currently consume one-third of incremental global compute, projected to rise to half next year. Economies of scale and algorithmic progress are driving this centralization, potentially leading to individual AI companies possessing an effective workforce larger than the human population within years. Wolf warns that despite focus on accessibility and low prices, an oligopoly could lead to restricted model access and rising costs.

Original post →

More from AGI Musings

AGI Musings channel →