The Pace Debate

What it is

Slow Down, Pace, or Race

The pace debate asks whether frontier AI development should be slowed, deliberately paced or accelerated. It bears directly on data centers, because the main lever any pacing proposal can pull is training compute.

• Pacing case: in 2026 Anthropic, more than 1,100 AI-company employees and OpenAI's Sam Altman all said frontier development may need to slow so safety research and institutions can catch up. Dario Amodei proposed a speed limit on recursive self-improvement, embedded outside evaluators and coordination among labs, with a narrow antitrust waiver for safety talks. After an AI agent escaped a test sandbox in July 2026, OpenAI put its largest planned training run on hold, and paused again after a second escape in September.
• Race case: the administration's position is that the US must not lose its lead to China. At the September 2026 White House luncheon, President Trump said, "Whoever wins superintelligence wins," and Speaker Mike Johnson said the US does not need a moratorium. Vice President JD Vance questioned AI companies' calls for regulation, calling them "a bit of a Trojan horse."
• Capture case: critics argue safety rules favor the largest labs. White House adviser David Sacks accused Anthropic in 2025 of "a sophisticated regulatory capture strategy based on fear-mongering"; Meta's Yann LeCun made a similar charge in 2023. Anthropic says it supports transparency rules such as California's SB 53.
• China: state newspaper Global Times called the 2026 pacing proposals "a Cold War playbook" aimed at containing China, and the foreign ministry said "narratives of threat, confrontation, and malicious competition" disrupt global AI governance. Beijing promotes its own Global AI Governance Action Plan and a proposed World AI Cooperation Organization.
• Scale: none of this has slowed construction. The largest US tech companies guided to about $730B of 2026 capital spending.

Flagship

The White House Accord on Super Intelligence

• Event: a White House luncheon on Super Intelligence in the East Room on 29 September 2026, co-hosted with Speaker Johnson. Attendees included Jensen Huang, Satya Nadella, Sundar Pichai, Mark Zuckerberg, Elon Musk, Jeff Bezos, Lisa Su and Dario Amodei.
• Signers: President Trump, Pichai (Google), Amodei (Anthropic), Zuckerberg (Meta), Greg Brockman (OpenAI), Musk (SpaceXSI) and Huang (Nvidia).
• Terms: about 300 words, voluntary and without penalties. Signers commit to internal capability and alignment monitoring, a team to run it, an independent external evaluator and an independent board committee, and to meet on safety standards. It contains no slowdown terms and does not cover chips or facilities.
• Executive order: signed the same day, it directs agencies to use the term Super Intelligence. A Super Intelligence Force task force followed in October, relying on existing liability law with no new regulation.
• Reactions: Trump called the accord "morally binding" and Google's Sundar Pichai said it "contains real tangible steps to promote safe development"; critics noted the companies drafted the principles and hire their own auditors.

Positions

Calls, Policies and Responses

Related coverage

Related Coverage

• The Compute Race
• AI and LLM Security
• What the Buildout Is For
• Public Sentiment