Nvidia's share of China's data centre AI chip market has fallen to 34%. On October 13, 2026, Reuters, citing a report by research firm IDC, published the first-half figures: Nvidia shipped about 678,000 AI accelerators in China from January to June, down 44% year on year, while Huawei shipped about 572,000 units — nearly double the 300,000-plus of a year earlier — for a 28% share. The once-unassailable leader and its closest challenger are now just six percentage points apart.
Where the rest of a 2-million-unit market stands
Total AI accelerator shipments in China reached about 2.008 million units in the first half. Behind Nvidia and Huawei, Alibaba's chip unit T-Head held 7%, Cambricon 5%, Baidu's Kunlunxin 4% and Hygon 3%. Nine of the 11 Chinese chipmakers tracked by IDC grew shipments from a year earlier. The category split is just as telling: GPU shipments — the chips most widely used to train large models — fell 26% to about 991,000 units. Demand did not disappear; the supply that can legally be sold into mainland China stayed constrained, and buyers moved their orders elsewhere.
Export controls rewrote the share table, point by point
US export controls restrict sales of Nvidia's most powerful AI chips to China and have narrowed the range it can offer, while Beijing promotes domestic alternatives and Chinese companies need approval to import compliant Nvidia chips. Nvidia CEO Jensen Huang has repeatedly criticised the curbs, arguing the policy hands China's roughly $50 billion AI chip market to Huawei. One caveat is worth keeping: IDC did not specify which products make up the 34%, or whether export-barred B200 and B300 chips that reached China through other channels were counted. Nvidia did not respond to a request for comment.
Procurement habits, once formed, are harder to reverse than one product cycle
The weight of these numbers is not how much share Nvidia lost, but that domestic substitution is becoming the default option for many buyers rather than a backup plan. Share is counted in units, which is not the same as per-chip performance or the high-end training market flipping overnight — but real deployments generate feedback, software adaptation and operational experience that feed the next procurement round. Globally, cloud providers' in-house chips are eating into the same market; the Maia 300 launch covered earlier on this site is Microsoft moving inference workloads onto its own silicon. Two variables are worth watching over the next six months: how much supply the approval process for compliant products actually releases, and whether Chinese vendors can sustain their growth rates as the base gets bigger.
