Anthropic has issued a stark warning regarding the escalating competition in artificial intelligence between the United States and China. Their newly released policy paper outlines a scenario where the next few years are critical for determining whether democratic nations can sustain their technological edge or allow authoritarian regimes to shape the future of AI. By 2028, the US must secure its computing advantage or risk ceding control to countries less committed to democratic values.
At the heart of Anthropic's argument is computing power, particularly access to advanced AI chips, which they identify as the main bottleneck in AI development. Current US export controls, along with advancements from companies like Nvidia, TSMC, and ASML, have positioned democratic nations favorably. The paper predicts that Huawei's compute capacity will only reach about four percent of Nvidia's total by 2026, decreasing further to two percent by 2027.
Nevertheless, challenges loom. Chinese AI labs are reportedly narrowing the gap, driven by two main factors: the illegal smuggling of chips and the use of foreign data centers to access US computing resources. Systematic distillation attacks have emerged, where thousands of fraudulent accounts harvest outputs from leading American models, allowing Chinese labs to replicate advanced capabilities. Anthropic has previously accused certain entities of generating millions of interactions with their AI model, Claude, through deceptive methods.
Anthropic’s paper presents two potential futures for AI development based on actions taken by US policymakers. In the first scenario, closing loopholes in chip exports and cloud access, along with legislative measures against distillation attacks, could allow the US to extend its lead in model intelligence by 12 to 24 months. This would establish American AI infrastructure as the backbone of the global economy, with norms and standards shaped by democratic governance.
On the other hand, if current loopholes remain unaddressed, Chinese AI labs could match their American counterparts. This scenario anticipates an increase in AI-powered surveillance and repression from Beijing, alongside a rise in global market share for Huawei's data centers, which could provide cheaper yet capable models. Anthropic cautions that such a competitive environment could compromise safety standards in AI development; only three out of thirteen leading Chinese labs have published safety evaluations to date, raising concerns about accountability and ethics.
The timing of this paper's release is particularly significant, coinciding with important political events, including discussions in Congress about AI export controls and chip licensing. Anthropic advocates for a restrictive and security-focused approach. Their message to policymakers is clear: decisive action now can secure a long-term advantage in AI technology leading up to 2028. Inaction risks enabling the growth of China’s AI capabilities, which could ultimately shape the global AI landscape in favor of authoritarian principles.
As competition intensifies, the stakes have never been higher. The next few years will likely define the trajectory of AI development and governance. The choices made by US stakeholders during this critical period will determine whether they can maintain a leadership position or if the future of AI will be shaped by those with fundamentally different values.
