Bernstein: AI Slowdown Talk Heats Up, Compute Spending Plans Remain Unchanged

TechFlow reports, according to Chaoxiang Research, a September 14, 2026 research note from Bernstein highlighted that Anthropic CEO Dario Amodei proposed a "pace framework," calling for a slowdown in the advancement of AI model capabilities through three steps: third-party evaluations, US regulation encompassing all AI firms, and coordination with China. The semiconductor sector is up 67% year-to-date, off 19% from its June high, and has rebounded 13% from its July low.

Bernstein notes that a slowdown in pace does not mean a slowdown in expenditure; Amodei is discussing capping training compute rather than stopping training altogether, advising a shift from an extremely rapid pace to a moderately fast one. AI semiconductor demand is increasingly driven by inference, and current compute capacity falls far short of meeting the needs of existing models. Concerns over China may be the primary driver behind the proposal, with Amodei advocating to maintain export controls on chips and equipment to China. Bernstein maintains an optimistic outlook on AI infrastructure growth, setting price targets of $400 for Nvidia, $575 for Broadcom, and $700 for Applied Materials, all assigned an Outperform rating.