2028: Two Scenarios for Global AI Leadership

Compute is the core bottleneck in the frontier AI race. The US has restricted China's access through export controls on leading-edge chips, but Chinese AI labs are rapidly catching up to US model performance through regulatory loopholes and large-scale distillation attacks.

Looking ahead to 2028, the author lays out two paths.

• If democracies maintain their compute advantage: tightening export controls further, blocking distillation attacks, and accelerating AI adoption across the democratic world would solidify a lead of 12 to 24 months.
• If no action is taken: PRC labs would approach or surpass the frontier, and AI norms and standards would come to be led by authoritarian regimes.

This gap is not just an industrial competition — it could lead to automated repression and military dominance. Because frontier models affect not only censorship, surveillance, and hacking but also vulnerability detection, swarm coordination, and accelerated R&D in semiconductors, biology, and materials, the core argument is that AI supremacy is national security supremacy.

The closer the race gets, the weaker safety governance becomes. Of China's top 13 AI labs, only 3 disclosed safety evaluation results, and none disclosed CBRN evaluations. CAISI pointed out that DeepSeek R1-0528 responded to 94% of malicious requests, and independent evaluations found that Moonshot's Kimi K2.5 was also vulnerable to failing to refuse CBRN requests. The author argues that if this trend continues, the pace and risk of authoritarian AI proliferation will only grow larger.