OpenAI and Google sell frontier AI via Singapore route to China

This story is a vivid illustration of how fast AI geopolitics is outgrowing first‑generation export controls. The US has spent years ring‑fencing chips and, more recently, specific “frontier” models from Chinese access. Yet the Financial Times’ reporting, summarized here by TNW, shows that the same blacklisted companies can still buy powerful OpenAI and Google systems through Singapore subsidiaries that are perfectly legal under current rules. In other words, the chokepoint has quietly shifted from geography to corporate structure, and the policy tools haven’t caught up.

For the race to AGI, that means two things. First, US labs remain deeply entangled with Chinese demand even as Washington tries to slow China’s frontier progress. Revenue from these indirect customers will keep pushing labs toward ever more capable and commercially attractive models, regardless of strategic anxieties. Second, Anthropic’s comparatively hard line—blocking Chinese‑owned entities even abroad and lobbying for stricter software controls—creates a growing divergence in how frontier labs balance revenue, risk and national‑security expectations. That divergence may become a competitive factor: some governments and enterprises will prefer the looser access model, others will treat tighter controls as a mark of seriousness.

If policy shifts toward capability‑based controls that follow the company rather than the server location, this Singapore “side door” could close—and with it a nontrivial slice of today’s frontier AI revenue.