Musk Backs Call to Slow Frontier AI: Why Timing and Computing Power Matter
Elon Musk publicly backed Anthropic CEO Dario Amodei's call for frontier labs to slow development, posting "Dario is right" on Sept. 12, 2026. X
Amodei had called on artificial intelligence companies to deliberately slow the pace of frontier model development. Frontier labs means the small group building the most advanced models. Anthropic cited rapid advancements and the potential for recursive self-improvement, where AI systems help design even more capable AI, as reasons for restraint. The Wall Street Journal
Amodei said artificial intelligence will surpass human cognitive abilities across most tasks within a "small number of years." Reuters
What that short timeline means for money is direct. Discounting is how investors value future profits in today's dollars. If big payoffs and disruptions arrive in a few years, they land inside a normal capex planning cycle, the usual few-year window for big business spending, not in a distant terminal value far in the future.
Amodei tied that speed to compute, or raw computing power. In prepared remarks for the AI Safety Summit, he said rapid progress is driven by available computation growing by 8x per year and unlikely to slow down. Anthropic
The practical read on that claim is simple. Talent and algorithms matter, but buying and building enough computing capacity sets the pace. When compute is the limit, spending power decides who moves fast.
Amodei leads Anthropic. Anthropic The company has published his statement titled 'Dario Amodei on American AI leadership' and, on July 27, 2026, his post titled 'Our position on open-weights models.' It has also published 'Policy on the AI Exponential' sharing two policy proposals to prepare for AI progress. Musk was named to TIME's 100 Most Influential People in AI for 2026. TIME
The broader context here is a coordination problem markets handle badly. A voluntary slowdown is a collective-action setup. Each lab gets some safety and reputational benefit from restraint but pays the full cost in lost capabilities, partnerships and unused computing capacity. Support from a rival lab head makes it easier to speak up. It does not create enforcement.
In my view, investors should separate the proposal from the probability. The proposal is to slow frontier training and deployment until governance catches up. The probability turns on verifiable commitments, shared evaluation standards and penalties for breaking ranks. None of those tools are described in the verified statements. Without them, markets will keep pricing continued expansion paid for out of big balance sheets.
For savers holding stocks or bonds, there is a tension to watch. If usable computing power keeps rising 8-fold a year, unused capacity costs money to sit on. That pushes companies to use it. Tighter limits on sharing frontier model weights help incumbents with closed products and business customers. Wider sharing cuts their edge but spreads use. Bond and stock desks will read the same safety talk as possible regulation, good or bad depending on time horizon and business model.
