Anthropic is warning that the race to advanced AI is moving into a phase that could reshape global power far beyond the tech industry. In its view, the country that controls the strongest AI systems will not just gain commercial advantage, but also influence over the rules, norms, and security balance of the next era.
The company places 2028 at the center of that warning. In a policy paper titled “2028: Two Scenarios for Global AI Leadership,” Anthropic argues that progress in frontier models is moving quickly enough to raise the possibility of human-level AI by then.
A future that looks like a data center full of geniuses
Anthropic describes the coming capability shift in striking terms, calling it a “nation of geniuses in a data center.” The phrase reflects its view that AI will no longer stop at answering questions or generating text.
Instead, these systems could actively accelerate scientific discovery, software development, cybersecurity work, and AI research itself. That kind of capability, the company says, would give the country behind it extraordinary leverage in economics, foreign policy, and defense.
Compute is the key battleground
The company sees compute as the single most important resource in frontier AI development. That makes advanced chips and the infrastructure needed to train large models the main strategic assets in the contest.
Anthropic says the United States still holds a meaningful lead because of its compute advantage. It also credits export controls under successive US administrations with helping preserve distance from China in advanced AI.
But the company warns that the gap could narrow quickly if loopholes remain open. It points to risks tied to chip export enforcement, overseas data centers, and access to models outside the United States.
How China could stay close to the frontier
Anthropic also says Chinese AI companies may still remain competitive through gaps in export controls. It highlights foreign compute access and what it calls “large-scale distillation attacks,” a method that uses outputs from leading US models to build similar capabilities at lower cost.
The company argues that Washington needs to limit those practices if it wants to keep its lead. Without tighter protection, it says, rivals may continue to benefit from US model output even when direct access to top hardware is restricted.
Two very different paths toward 2028
In the paper, Anthropic sketches two possible outcomes. In the first, the US and its allies tighten control over advanced chips, stop smuggling, and speed up domestic AI adoption.
Under that scenario, the company says democratic countries could hold a strong advantage. US systems could remain 12 to 24 months ahead of Chinese rivals in intelligence and capability.
The second scenario is far more troubling for Washington. If enforcement stays weak, foreign access to compute remains open, and US systems continue to be used for research and development, Anthropic says China could stay very close to the AI frontier.
That outcome would also allow China to help shape the global AI ecosystem. Anthropic warns that this would widen the use of AI in surveillance, cyber operations, and military applications.
Pressure for tighter policy
Anthropic is pushing for stricter limits on smuggled chips, overseas compute access, and semiconductor manufacturing equipment. It also wants stronger safeguards against distillation attacks aimed at US AI models.
At the same time, the company says dialogue with AI researchers in China on safety should continue. Even so, it argues that such engagement works best when the US still has a clear technological lead.
By treating 2028 as a critical turning point, Anthropic is signaling that the AI race has already moved beyond its early stage. Decisions made now on chips, infrastructure, model access, and enforcement may help determine who sets the terms of the global AI order in the near future.
