Six days after Washington cleared sales of Nvidia's H200 AI chips to roughly ten Chinese firms, no published purchase order has followed, leaving the paper's permission-without-procurement frame intact for another beat.

Reuters reported the clearance on May 14, framed by Jensen Huang's Beijing visit as a "breakthrough" moment for the company. [1] The clearance itself is a real document. What has not arrived is its counterpart on the buyer side: a Chinese firm naming volumes, timing, or terms, or a state planner blessing dependence on a U.S. accelerator the export regime can revoke at the next political turn.

That absence is not a passive fact. The U.S. Trade Representative said Chinese firms will make their own determinations, and Beijing remains committed to domestic alternatives. The same regulator who unlocks a chip can lock it again, and a Chinese hyperscaler buying H200s today is a Chinese hyperscaler whose roadmap is hostage to next quarter's bilateral weather.

There is also a quieter signal. Domestic Chinese accelerator suppliers have been told for two years that the political ceiling on imported compute is the operational floor for their own products. A flood of H200 orders would partly undo that signal. A pause is consistent with state-planning preference even if Huang's invoice book disagrees.

The paper's discipline is to treat permission, demand, and procurement as three different artifacts. The first is on the record. The third is not. Until a Chinese buyer publishes an order, the H200 story is still about what the United States allows, not about what China has decided to buy.

-- DAVID CHEN, Beijing