OpenAI does not officially offer its services in mainland China, Hong Kong or Macau, yet it is hiring Mandarin-speaking developer experience engineers in Singapore. On May 14, the day US President Donald Trump met President Xi Jinping in Beijing, Anthropic published a policy paper arguing the United States and its allies must lock in a 12-to-24-month lead in frontier artificial intelligence (AI) by 2028 to avoid “authoritarian AI leadership”.

In the same quarter, Anthropic disclosed US$1.56 million in federal lobbying, a 333 per cent year-on-year increase, with “export controls” and “AI and national security” among the listed topics.

Taken together, these facts reveal an industry whose stated posture and revealed behaviour are moving in opposite directions. American AI companies are arguing for tighter controls on China while positioning themselves around China-adjacent markets those controls might restrict. They frame China as a strategic threat while hiring to build the language and developer ecosystems they cannot officially serve.

The contradiction is not simply corporate inconsistency. It is the shape of an industry caught between two AI contests that reward opposite behaviour.

The first is over frontier capability: who builds the most powerful models and who controls the compute needed to build them. Anthropic’s memo articulates this logic. Compute is the dominant input. The US still holds a lead, and that lead could compound because stronger models help build stronger successors. The prescription is tighter export controls, hardened access to American systems and faster adoption among allies.

In this contest, AI is a strategic asset whose value lies in scarcity. The argument rests on a theory of technological leadership: that early advantage can be defended long enough to lock in standards, alliances and institutional norms.