Who’s Afraid of Chinese Models?

The rise of Chinese open-weights AI models, such as Kimi K3 and Quan 3.8 Max, challenges the dominance of U.S. frontier labs by shifting the focus from training costs to the marginal costs of inference. While tokens are often misidentified as a commodity, true fungibility lies in the intelligence generated, where Chinese labs gain a structural advantage through state-supported openness and the distillation of Western frontier models. Current U.S. policies, which restrict access to powerful models for cybersecurity, inadvertently force domestic companies to rely on Chinese alternatives for critical infrastructure defense. To maintain competitiveness, the U.S. must move beyond protectionist guardrails, clarify that data collection for training is fair use, and ensure domestic open-weights developers can compete on a level playing field, ultimately allowing frontier labs to win through superior performance rather than regulatory exclusion.