China's AI Chips and Export Controls: Ascend, SMIC — 2026-10-15

Top developments

IDC Report: Nvidia Share Drops to 34%, Huawei Hits 28%

Recent IDC data cited by Reuters shows Nvidia’s share of China’s data center AI chip market fell to 34% in the first half of 2026, down from previous highs due to US export controls. Huawei captured 28% of the market, narrowing the gap to just six percentage points. Nvidia’s shipments fell by 44% year-over-year, while Huawei’s shipments nearly doubled to reach 572,000 units compared to Nvidia’s 678,000 units.

Non-GPU Accelerators Outpace GPUs for First Time

For the first time, shipments of non-GPU accelerators in China exceeded those of GPUs in the first half of 2026. This trend is driven by the adoption of domestic chips from Huawei, Alibaba’s T-Head, Baidu’s Kunlun Chip, and Cambricon, which collectively hold 95% of the non-GPU market. IDC projects that total AI accelerator shipments in China will grow by 19% in 2026, while GPU shipments are expected to decline by 9%.

Huawei Claims Ascend Sales Top Nvidia; Supply Still Constrained

Huawei Executive Director Eric Xu stated that according to internal Huawei data, Ascend chip sales have already surpassed Nvidia’s in the Chinese market. However, he acknowledged that supply remains far below domestic demand. Despite this, top AI labs like DeepSeek continue to rely on Nvidia hardware and CUDA for model training, highlighting the persistent software ecosystem gap.

SMIC Capacity Allocation: Huawei Dominates, Cambricon Gains

Huawei has secured approximately 43% of SMIC’s N+2 advanced node capacity for 2026, amounting to roughly 15,000 wafers per month under a five-year agreement. Cambricon has secured an estimated 9% to 11% of this capacity. The remaining capacity is contested by other players like Biren, Moore Threads, and T-Head. This allocation is critical as SMIC remains the primary foundry for domestic AI chip production due to sanctions limiting access to TSMC.

Local view

Chinese media outlets such as Huxiu and Sina Finance are highlighting the "marathon" nature of the domestic chip race, noting that while Huawei leads, the ecosystem is still maturing. Reports emphasize that government agencies, internet companies, and telecom operators are increasingly adopting domestic products like Ascend and Cambricon. Local analysts point out that while performance gaps remain compared to Nvidia’s latest offerings, the reliability of supply and political pressure are driving the shift.

Context & numbers

• Market Share (H1 2026): Nvidia: 34%; Huawei: 28%.
• Shipments (H1 2026): Nvidia: 678,000 units; Huawei: 572,000 units.
• SMIC Capacity: Huawei holds ~43% of N+2 allocation; Cambricon holds ~9-11%.
• Growth Projections: IDC expects total China AI accelerator shipments to grow 19% in 2026, with GPU shipments declining 9%.

On the radar

• Shenzhen APEC Planning: Preparations for the November Trump-Xi meeting in Shenzhen are ongoing, with potential implications for future trade and technology policies.
• DeepSeek Adaptation: Continued monitoring of how major Chinese AI labs like DeepSeek adapt their models to run efficiently on Huawei Ascend hardware without relying on CUDA.
• Export Control Enforcement: Watch for further BIS guidance on enforcement of license requirements for advanced computing items, particularly regarding transshipment via Southeast Asia.