Frontier AI companies probably can't leave the US

A new analysis from Redwood Research, authored by Anders Woodruff, concludes that leading 'frontier' AI companies like Anthropic and OpenAI are effectively trapped in the United States by existing legal authorities, regardless of future political or regulatory pressures. The report argues that even if companies become deeply unhappy with domestic politics, democratic backsliding, or restrictive federal AI regulations, the US executive branch can easily prevent them from relocating their core assets—specifically their advanced AI chips (GPUs), capital, and intellectual property—using powers already on the books. This reality fundamentally alters the strategic landscape for AI governance, removing a potential threat companies have used in the past to negotiate against stringent rules.

The technical mechanism hinges on two key laws: the Export Administration Regulations (EAR), which classify advanced AI chips as controlled items, and the International Emergency Economic Powers Act (IEEPA). The president can use EAR to block the physical export of thousands of chips, as moving them abroad constitutes an export. Simultaneously, IEEPA provides authority to freeze the financial transactions necessary for offshoring operations. The analysis cites the precedent of export controls against Huawei and notes the severe reaction Anthropic recently faced merely for disputing Pentagon terms, suggesting a full departure would trigger a massive political and national security response. Consequently, AI companies lose a major leverage point, ensuring the US government retains decisive control over the geographic and strategic future of frontier AI development.