Nvidia's grip on China's data-center AI chip market loosened sharply in the first half of 2026. According to an IDC report seen by Reuters, the US chipmaker shipped 678,000 AI accelerators into China between January and June, down 44% from a year earlier, leaving it with 34% of the market — just six points ahead of Huawei at 28%.
Huawei shipped 572,000 accelerators over the same period, up from just over 300,000 a year earlier, and has become the anchor of the domestic replacement push. The rest of the leaderboard: Alibaba's chip-design unit T-Head at 7%, Cambricon at 5%, Baidu's Kunlunxin at 4% and Hygon at 3%. Nine of the eleven Chinese chipmakers IDC tracks grew shipments year over year.
The deeper structural story is what kind of chips are being bought. Total AI accelerator shipments in China reached 2.008 million units in the half. GPU shipments — the workhorses of AI training — fell 26% to 991,000 units, as Washington's export controls kept the most advanced parts off the market and Beijing limited approvals for the compliant models Nvidia is allowed to sell. Shipments of other AI accelerators built for specific inference workloads jumped nearly 80% to 1.016 million units, overtaking GPUs by volume for the first time. Huawei, T-Head, Kunlunxin and Cambricon together account for 95% of that non-GPU segment, with adoption concentrated in government agencies, internet companies and telecom operators.
IDC expects the mix to keep shifting: total accelerator shipments in China should grow 19% this year even as GPU shipments fall 9%. The firm also argues that competition is moving beyond the performance of a single chip toward system-level capability — how well large clusters behave together, a dimension where Huawei's Ascend ecosystem and its software stack have been investing heavily.
Nvidia did not respond to a request for comment on the report. CEO Jensen Huang has repeatedly criticized the export-control regime, arguing that it forced a US company to hand what he sizes as a $50 billion China AI chip market to Huawei. IDC did not specify which Nvidia products made up the 34% share, or whether smuggled B200- and B300-class parts are reflected in the figures.
The trend line matters more than any single half. Every quarter that Chinese buyers build around domestic accelerators, the switching costs of coming back rise — in software, in cluster designs, in procurement habits that now assume a second source. For Nvidia, the $50 billion question is not whether China's AI buildout continues, but whether any of it still runs on its silicon when the rules eventually loosen.
Comments (0)
Log in to join the discussion
Log InNo comments yet