The instruction was relayed through Nvidia's sales force and regional teams. No written document backs it. Nvidia neither confirmed nor denied the move when AFP approached it.

The 51 percent non-Chinese floor applies only to new contracts. Existing agreements are untouched, according to five people cited by the FT. The unit of measurement is unspecified: the figure could track revenue, GPU hours, customer count, or contracted capacity.

The constraint targets the cloud provider's customer mix, not the chip's destination. Chinese model teams that rent compute from providers outside China are the channel export controls miss, and this instruction closes it at the commercial layer.

Washington has limited sales of Nvidia's most advanced chips to China since 2022, but the hardware continues to enter the country. The reported restriction adds a second wall for Chinese AI developers after the chip embargo.

What remains unmeasured: how the figure is audited, whether existing contracts are truly untouched in practice, and whether a verbal instruction survives contact with a partner whose revenue depends on Chinese workloads.

Leads with the absence of a written policy rather than the number, which keeps the piece distinct from the earlier brief on the same rule. The unit ambiguity is surfaced early so the reader does not assume the 51 percent is revenue-based. Closes on the audit question rather than restating the mechanism.