Huawei has trained Pangu Pro, an artificial intelligence model with 505 billion parameters, without using Nvidia chips, according to a report by Tech Times carried on Google News.
Parameters are the internal settings a model adjusts as it learns; the count is a rough proxy for scale and capacity. A 505-billion-parameter model sits in the same broad size class as the large frontier systems built by American labs, and training one requires enormous quantities of specialized computing hardware — the market Nvidia has dominated.
That is what makes the claim significant. U.S. export controls have restricted Chinese access to Nvidia's most capable AI processors, and a Chinese company demonstrating that it can train at this scale on domestic alternatives would suggest those restrictions bind less tightly than intended.
But Tech Times frames its report with a caveat in the headline itself: the supply chain "tells a different story." In other words, the outlet suggests the picture behind the announcement is more complicated than a clean break from Nvidia. The available source item does not spell out what that complication is, and the specifics — which chips were used, where they were fabricated, and how the training run was verified — are not established in the material at hand.
Why it matters: if Chinese firms can train frontier-scale AI without Nvidia, the chip export controls that anchor U.S. technology policy lose much of their leverage — but a headline claim is not the same as a verified independent supply chain, and the difference is the whole story.