Huawei's Q&A says a SuperCluster with 256,000 computing cards is currently being deployed and tested. The Register's wire piece says the same company has crossed a market-share line, quoting a chairman who says "Ascend has surpassed Nvidia." Those two pages are not measuring the same thing. The AP Multimedia Newsroom document is the one the desk trusts, and the 256,000-card number decides it.

The weaker page is the wire article, because it turns a deployment claim into a sales claim. The claim is short, but it carries no denominator. The trusted page anchors the story to a physical cluster that can be checked against power, rack, and card counts.

The desk's judgment is that Huawei is constrained by capacity, not by proof of market share. The primary release proves a deployment scale, not a sales ranking. It does not prove that Ascend units outsold Nvidia units in China, because no shipment count, revenue split, or customer list appears. For the buyer, the useful fact is that a 256,000-card cluster is in test, which changes procurement risk more than a verbal claim of supremacy.

What remains unmeasured is the actual share of Chinese AI accelerator purchases. The release omits Nvidia shipment volumes, Huawei revenue, and the number of deployed cards outside the 256,000-card cluster. A separate gap is whether the one-million-processor architecture has been built, only that it is designed to scale.

Huawei is constrained by capacity, not by proof of market share. The 256,000-card cluster changes procurement risk more than a verbal claim of supremacy.

The document leaves out Nvidia shipment volumes, Huawei revenue, and the number of deployed cards outside the 256,000-card cluster.

The measurable fact is a 256,000-card SuperCluster in deployment and test, not a sales ranking. Buyers should treat the market-share claim as unverified until shipment or revenue data appears.