Nvidia's China AI-chip share fell to 34% — but the shipment mix is the real story

Nvidia still leads China's data-centre AI chip market. That's the most generous way to read the new IDC figures reported by Reuters — and from there, the numbers get worse for Santa Clara.

In the first half of 2026, Nvidia shipped 678,000 AI accelerators in China, down 44% from a year earlier, leaving it with 34% of the market. Huawei nearly doubled its shipments to 572,000 units over the same period, good for a 28% share. Alibaba's T-Head (7%), Cambricon (5%), Baidu's Kunlunxin (4%), and Hygon (3%) round out the field — and nine of the eleven domestic chipmakers IDC tracks grew year over year.

The deeper story is in the shipment mix. GPU shipments in China fell 26% to 991,000 units, while shipments of non-GPU AI accelerators — chips built for specific AI tasks like running trained models — rose nearly 80% to 1.016 million, overtaking GPUs by volume for the first time. Huawei, T-Head, Kunlunxin, and Cambricon together account for 95% of that non-GPU segment, which IDC says is being adopted by government agencies, internet companies, and telecom providers. IDC expects the trend to hold: total Chinese accelerator shipments up 19% this year even as GPU shipments fall 9%.

That's export controls working in reverse. Washington restricted Nvidia's most powerful chips to starve China's AI ambitions of compute. What starved instead was Nvidia's own China business, while a domestic inference-hardware industry grew up around the gap. Jensen Huang has said the policy forced a US company to "cede China's $50 billion AI chip market to Huawei." The IDC numbers give that complaint a receipt.

It's also worth noting how far ahead of the data some narratives are running. Last week Huawei claimed, using its own figures, that Ascend chips had already overtaken Nvidia's China sales by roughly 50% to 8%. IDC's neutral count puts the race much closer — 34% to 28%. Either way, the direction is the same.

IDC's read on where the competition goes next is the part Nvidia should worry about most: away from single-chip performance and toward system-level capability — how well chips work together in a cluster. That's Huawei's home turf, where tight integration of Ascend silicon with its own networking and cloud stack is the pitch. Huawei has two more AI chips arriving in 2027, per Reuters. Watch Thursday's TSMC earnings for the next data point on who the world's foundries are really building for.