Bernstein: AI Slowdown Talk Heats Up, Compute Spending Plans Remain Unchanged
According to TechFlow Research, a September 14, 2026 report from Bernstein states that Anthropic CEO Dario Amodei has proposed a "pace framework," advocating for a slowdown in the rate of AI model capability advancement, encompassing three steps: third-party evaluations, US regulation covering all AI companies, and coordination with China. The semiconductor sector is up 67% year-to-date, down 19% from its June high, and rebounded 13% from its July low.
Bernstein believes that a slowdown in pace does not mean a slowdown in spending; Dario is referring to capping training compute, not halting training altogether, suggesting a shift from extremely fast to moderately fast. AI semiconductor demand is increasingly driven by inference, and current compute capacity falls far short of meeting the requirements of existing models. Concerns regarding China may be the larger driver behind the proposal, with Dario recommending the maintenance of bans on chip and equipment exports to China. Bernstein maintains a positive outlook on the AI buildout, setting price targets of $400 for NVIDIA, $575 for Broadcom, and $700 for Applied Materials, all with an Outperform rating.