Nvidia and Huawei 6 Points Apart: Shifting Competitive Landscape in China's AI Chip Market

Nvidia's shipments of AI accelerators to China fell 44% year on year to 678,000 units in the first half of 2026, while Huawei shipped 572,000 units, nearly double the prior year. Jensen Huang has criticized US export restrictions for handing China's roughly $50 billion AI chip market to Huawei. In its first-quarter earnings call, Nvidia's CFO said no Hopper-architecture data center products were shipped to mainland China that quarter, against $4.6 billion in revenue from this business before the restrictions. The company booked $4.5 billion in inventory impairments and order cancellation losses and expects an $8 billion revenue loss in the second quarter from H20 sales limits. Huang said China's share of revenue had fallen to about 9%, down from at least one-fifth of data center revenue three years earlier. Huawei's growth has been constrained by supply. The Ascend 910C shipped about 812,000 units in 2025, taking nearly half the domestic market, and delivered 150,000 more in the first quarter of 2026 to 37 leading large-model companies. The bottlenecks are SMIC's second-generation 7nm process yield and high-bandwidth memory supply. Huawei plans to raise 910C output to about 600,000 units in 2026 and target 1.6 million Ascend wafers in total. Non-GPU accelerators grew nearly 80% to 1.016 million units in the first half of 2026, surpassing GPU shipments (991,000 units) for the first time. Huawei, T-Head, Kunlunxin and Cambricon hold 95% of this segment. Huawei's CloudMatrix 384 links 384 Ascend cards into one supernode, reflecting a shift from single-chip performance to system-level competition. Huawei has outlined a three-year roadmap: Ascend 950PR in Q1 2026, 950DT in Q4 2026, 960 in 2027 and 970 in 2028. The 950PR will carry 112GB of in-house HBM with 1.4TB/s bandwidth.