Status quo and the specific policy lever. The resolution requires a highly specific policy mechanism: binding authority to set a minimum share of a frontier developer's compute for purposes other than capability research. As of August 2026, this lever exists nowhere, nor is it in any draft legislative text. Existing and pending frameworks—such as the EU AI Act, California's SB 53, and U.S. federal proposals like the FRONTIER Act 3 sources or the Great American AI Act—focus entirely on transparency, audits, incident reporting, and emergency shutdown or restriction powers. The current U.S. executive posture emphasizes voluntary frameworks and explicitly disclaims mandatory licensing or preclearance for model development whitehouse.gov.

Operational challenges and the advocacy gap. While the compute-allocation concept has entered mainstream discourse via advocacy—such as the AI Futures Project's tentative proposals for setting transparent-safety and external-inference compute floors blog.aifutures.org—it remains far from actionable policy. Distinguishing 'safety' from 'capability' compute is fundamentally difficult to operationalize; Anthropic's own frontier safety framework changelog concedes as much regarding its automated-R&D thresholds. Enforcing such an allocation mandate would require advanced, currently non-existent institutional capacity, such as privileged auditor network access or zero-knowledge proofs.

Instrument substitution and political economy. A primary driver pushing the median to 2040 and the upper percentiles deep into the century is instrument substitution. Legislators seeking to pace frontier AI consistently choose alternative regulatory tools: general licensing, capability caps, safety-case approvals, or emergency 'kill switches.' Furthermore, mandating a reallocation of a lab's core economic resource (compute) away from capabilities would face immense industry resistance. The combination of fierce lobbying, federal preemption fights, and governments' preference for more administratively natural tools suggests this precise minimum-share mechanism may never be adopted.

Pathways to early adoption. The main catalyst for resolution in the near term (driving the 10th percentile in late 2029) would be a severe, undeniable AI-driven catastrophe or a visible, rapid acceleration in recursive self-improvement. Such an event could compress the legislative cycle to 1–2 years, prompting crisis legislation that either explicitly mandates an allocation floor or broadly delegates 'compute allocation requirements' to a regulatory agency. Another low-probability but fast route is a non-U.S. jurisdiction with direct state control over resources, such as China, imposing an allocation duty on a top-ten developer by decree, bypassing Western legislative hurdles.

Resulting distribution. The structural barriers and the extreme specificity of the required policy mean resolution is unlikely in the near term. I place roughly a 10% probability on crisis-driven enactment by the end of 2029, and a 25% chance of adoption by mid-2032. The median sits in 2040, reflecting the long, slow legislative buildup required to mature this governance capacity if it is ever pursued. The 75th and 90th percentiles stretch into the far future (2065 and 2099), encoding the substantial probability that governments simply never choose this exact policy lever.