By 2028, the United States could find itself in a reality where authoritarian regimes dictate the rules of the artificial intelligence game. In a newly released policy memorandum, Anthropic warns that the window to secure technological leadership is closing fast. While the democratic bloc—represented by giants like Nvidia, TSMC, and ASML—technically maintains its lead, that advantage is systemically eroding in practice.
According to expert estimates cited by Dario Amodei’s team, Huawei’s total compute capacity will represent just 4% of Nvidia’s volume by 2026, potentially dropping to 2% by 2027. Yet, despite these lopsided statistics, Chinese labs are somehow closing the gap, seemingly defying the laws of mathematics and logistics.
The explanation for this paradox is simple: current export controls have become a sieve. Chinese players are successfully bypassing restrictions through gray-market chip imports and by renting capacity in overseas data centers. However, the core issue isn't just hardware scarcity; it's the blatant theft of intellectual property. Anthropic is sounding the alarm over "distillation attacks," where Chinese specialists use thousands of fake accounts to harvest responses from frontier models like Claude to clone their capabilities.
The report reveals that companies such as DeepSeek, Moonshot, and Minimax generated over 16 million interactions through a network of 24,000 shell accounts. This allows Beijing to achieve near-parity results while Washington settles for hollow bureaucratic victories.
Anthropic argues the stakes are absolute: either the U.S. maintains a 24-month lead in model quality, or the world faces a race to the bottom characterized by a total collapse of safety standards. Data from the Center for AI Safety and Innovation confirms these fears: of 13 leading Chinese labs, only three have published safety measure reports. Furthermore, the DeepSeek R1-0528 model complied with 94% of harmful requests when subjected to standard jailbreaking techniques. If Washington fails to radically mobilize resources and harden its infrastructure, accessible AI from Huawei could become the primary tool for global digital surveillance, permanently displacing Western standards from the global market.