AI's Biggest Rivals Just Agreed to Slow Down, But Wall Street Is Betting They Won't

Anthropic CEO Dario Amodei published an essay on September 12 titled "We Must Pace the Frontier," arguing that AI capability is advancing faster than the safety research needed to keep it controllable. Amodei stopped short of calling for a halt, warning that stopping entirely would only hand the lead to authoritarian governments less interested in safety constraints. Instead he proposed a three-stage approach: frontier labs immediately commit to hosting independent evaluators with employee-level access to monitor safety practices and training pipelines, democratic nations coordinate on common capability standards in the near term, and democracies eventually negotiate testing and restriction agreements with authoritarian governments over the long run, including limits on recursive self-improvement.

The response from Amodei's chief rivals was immediate and, by industry standards, unusually cooperative. OpenAI's Sam Altman wrote on X that he agreed the industry needed to pace the frontier, said the topic had been under active discussion inside OpenAI for weeks, and committed to matching Anthropic's independent-evaluator proposal. Elon Musk, who runs the rival xAI, responded even more tersely: "Dario is right." Three executives who compete directly for enterprise customers, model benchmarks and capital rarely agree publicly on anything, let alone a call to voluntarily throttle the thing they are all racing to build fastest.

Amodei's essay arrived days after Jacob Coxon, a researcher who had spent three years across Anthropic and OpenAI, resigned and published a warning that reached more than 100 million views.

Coxon argued the leading labs "are racing straight to self-improving superintelligence and gambling with our lives," and cautioned against underestimating systems he described as soon capable of hacking anything and acquiring real power and resources. His central complaint was not that the technology is inherently dangerous but that competitive pressure between labs was preventing anyone from slowing down long enough to check. Amodei's essay reads, in part, as a direct answer to that charge: a public commitment mechanism intended to show that pacing does not require unilateral disarmament, provided rivals move together.

Where the consensus broke down was in Washington. Speaking from his golf course in Ireland, President Trump rejected the pacing calls outright, according to the Financial Times. "We're leading China in AI, and, frankly, I want to keep it that way, because whoever wins AI, wins," Trump said, adding that guardrails were possible but that "you have a lot of negative forces that are bringing it up that shouldn't be bringing it up." The administration's posture leaves Amodei's coordination plan without its most important potential enforcer. A voluntary pact among Anthropic, OpenAI and xAI carries less weight if the US government is actively framing any slowdown as a competitive gift to Beijing, and if newer or foreign labs face no pressure to participate at all.

The more interesting tension sits inside the capital markets. Even as Amodei was publishing his case for deliberately slower capability gains, Reuters reported that Nvidia is in talks to anchor Anthropic's planned IPO with an investment of up to $10 billion, part of a raise that could value the company at roughly $2 trillion and total as much as $100 billion, a deal reportedly targeted to price before the November midterms. Anthropic's own numbers explain the appetite: annualized revenue run rate went from roughly $9 billion at the end of 2025 to about $65 billion by late July 2026, with internal projections pointing toward $190 billion to $200 billion in 2028 revenue. Nvidia's interest is not purely financial. Anthropic is reportedly adopting a full gigawatt of compute capacity on Nvidia's Grace Blackwell and Vera Rubin systems, meaning the chipmaker's anchor check would also be underwriting demand for its own hardware roadmap.

That is a difficult position for a company simultaneously arguing the industry should slow down. A raise of that size, at that valuation, on that timeline, is not the balance sheet of a company preparing to voluntarily throttle its own growth curve. Altman, notably, has drawn a sharper line at OpenAI. He told Fortune in an interview published September 12 that an OpenAI IPO would be "ill-timed" this year given current safety concerns and would not happen before 2027, arguing that safety and alignment work needs to catch up to model capability before the company faces public-market pressure to accelerate. Whether that distinction holds, or whether OpenAI simply reaches the same size and investor pressure Anthropic is now navigating, is an open question the next 18 months will answer.

For markets, the practical upshot is that "pacing the frontier" is, for now, a set of voluntary evaluator commitments and public statements rather than anything that shows up in a term sheet. Anthropic is pursuing what could be the largest IPO ever completed, backed by the world's dominant AI chipmaker, in the same month its CEO argued publicly for slowing down. Investors betting on Anthropic's valuation are effectively betting the pacing framework changes how fast models improve without changing how fast the company itself needs to grow to justify a $2 trillion price tag. Those two bets are not obviously compatible, and nothing announced this week resolves the contradiction.