China's AI chip market shifts: Nvidia at 34%, Huawei reaches 28%

IDC data reported by Reuters shows Nvidia's share of China's data-center AI accelerator market falling to 34% while Huawei reaches 28%. What is changing?

Main report

On October 13, 2026, Reuters reported IDC data showing Nvidia held 34% of China's data-center AI accelerator market in the first half of 2026. Nvidia shipped about 678,000 accelerators, down 44% year over year. Huawei shipped roughly 572,000 units and reached a 28% share, narrowing the gap to six percentage points.

A market moving beyond GPUs alone

Total AI accelerator shipments in China reached approximately 2.008 million units in the first half. Non-GPU accelerators grew 80%, while GPU shipments fell 26% to about 991,000 units. IDC expects total accelerator shipments to grow 19% in 2026 even as GPU shipments may decline 9%.

These figures describe shipments and share in one market. They do not by themselves establish technical performance, profitability or global leadership.

Analysis

Why the Nvidia–Huawei gap narrowed

Reuters linked the shift partly to US export controls that restricted the availability of Nvidia's higher-end products in China. Nvidia's regulatory filings also identify export licensing requirements and restrictions as direct risks to its ability to serve the market. The resulting supply gap gave Huawei and other domestic suppliers room to expand their hardware and software ecosystems.

Competition is moving from chips to complete systems

Data-center customers compare more than single-chip speed. Cluster scaling, interconnect bandwidth, power use, developer tools, software reliability and total operating cost all matter. A platform can therefore gain share through availability and workload fit without leading every benchmark.

Growth in specialized accelerators also suggests buyers are matching architectures to particular AI workloads instead of assuming every task must run on a GPU.

What to watch

What businesses should take from the shift

AI infrastructure purchases should not be based on market share or brand recognition alone. Test real workloads and measure throughput, power, stability, software migration cost, supply availability and support quality. A cheaper system can become more expensive if software must be heavily rewritten or scaling is difficult.

For developers, a more diverse accelerator market increases the value of portable deployment, model optimization and inference-cost monitoring. This is technology and market analysis, not investment advice.

This article belongs to TBN Grow's AI infrastructure economics cluster. Read the pillar article on AI infrastructure's trillion-dollar returns test for the broader investment context.

Verified sources: Reuters, October 13, 2026, Nvidia's SEC filing and IDC's global AI infrastructure analysis.