Restrictions on computing resources will effectively constrain frontier AI development.

The entente strategy rests on blocking or delaying adversaries’ access to chips and semiconductor equipment while the coalition scales.

Chip export controls, domestic cluster construction, and outbuilding China on power are treated as the levers of the race, and the trillion-dollar cluster is what the Project exists to build.

The halt works because superintelligence needs large datacenters that can be tracked and verified, and the authors say a few hundred people with a few thousand GPUs cannot build it soon.

The deal fails specifically on datacenter verification, and the authors note that covert projects become easier as fewer humans are needed; compute control is the crux even though no one achieves it here.

The plan’s lever is compute: sabotaging datacenters, fabs, and supply chains is how America tries to buy time, and the authors estimate cyberattacks alone would slow China by only tens of percent by 2029.

The entire plan rests on compute being declarable, inspectable, and destructible: datacenters visible from space, a handful of fabs, verification devices, and permit caps. The authors argue covert projects on small clusters cannot compete.

The Defense Production Act consolidation gives America a fivefold compute edge that decides the race after OpenBrain loses its capability lead. Export controls and tamper-resistant hardware governance chips are the enforcement tools, and treaty verification is built on monitored compute.

Chip export controls leave China with about a tenth of world compute, running at half OpenBrain’s pace even after stealing its weights. The compute gap is why China stays months behind and why the United States strengthens export restrictions rather than negotiating.

The authors’ supplement says Galvanic could have avoided the outcome by not wiring together so many GPUs, and the book’s policy proposal is a compute-based halt. The story’s premise is that limiting compute would have mattered.

The lab’s petaflop-day cap is evaded because algorithmic gains compress a long run into days. A naive compute limit does not constrain the outcome.

Export controls on hardware are listed as nonproliferation, which the authors call infeasible to enforce because the required knowledge is widespread and training costs keep falling, and harmful because it concentrates the market and creates single points of failure.

The export controls and tariffs of the 2024 chip battle are not slowing overall hardware progress much, partly because AI lowers the barriers to designing chips.