Chinese Chipmakers Now Control 41% of China's AI Accelerator Market — Up From Near Zero in 2022

A market intelligence report published this week has confirmed what many in the semiconductor industry have long suspected: US export controls on advanced AI chips have, in a fundamental sense, worked — just not in the direction American policymakers intended. Chinese domestic chipmakers now control 41 percent of China's AI accelerator market, up from a position of near-irrelevance before the first major restrictions took effect in October 2022.

The Export Control Gamble

When the Biden administration imposed sweeping restrictions on the export of NVIDIA A100 and H100 chips to China in October 2022, the strategic logic was containment: deny China access to the hardware needed to train frontier AI models, thereby slowing the country's AI development relative to the United States. NVIDIA and other US chip companies objected loudly, warning that the controls would simply accelerate Chinese domestic alternatives. Three and a half years on, that warning appears to have been prescient.

The 41% market share figure represents chips primarily from Huawei's Ascend series, Cambricon, and a cluster of newer entrants backed by state investment. In raw terms, the domestic share of China's AI accelerator market has roughly doubled since 2023 — the year the export controls were tightened further and extended to cover lower-performance chips like the H800 and A800 that NVIDIA had specifically designed to thread compliance requirements.

The Huawei Factor

Huawei's Ascend 910B and 910C chips are now widely used in major Chinese AI training clusters, including those operated by Baidu, Alibaba Cloud, and several government-affiliated research institutions. The chips do not match NVIDIA's latest H200 or Blackwell architecture on raw performance benchmarks. But they have become "good enough" for a wide range of training workloads — and Chinese AI labs have become expert at scaling clusters of domestic chips to compensate for per-chip performance gaps.

ByteDance, Tencent, and Alibaba have all reported accelerating their adoption of domestically sourced AI accelerators, partly for geopolitical risk hedging and partly due to supply reliability concerns after NVIDIA's export-compliant products were themselves placed under tighter restrictions in late 2024.

What 41% Means for Policy

The market share data poses a direct challenge to the bipartisan consensus in Washington that export controls are an effective long-term AI containment strategy. The 41% figure suggests that China's domestic AI hardware industry has crossed a viability threshold — it is no longer an emergency response to export controls but a structurally competitive industry segment with its own manufacturing scale, software ecosystem, and customer base.

Critically, the Chinese government has invested an estimated $47 billion in semiconductor self-sufficiency through its "Big Fund" program since 2014, with a third injection reportedly in preparation. That level of state backing makes the trajectory difficult to reverse through incremental tightening of existing controls.

The remaining 59% of China's AI accelerator market is still served by NVIDIA through its export-compliant H20 chip — a product specifically designed to meet US licensing requirements. That slice of the market remains under active policy review, with the Biden administration having initiated a process to restrict H20 exports as well before leaving office, a process the current administration has not formally terminated.

The Global Implication

The 41% figure is not just a China story. It signals the emergence of a bifurcated global AI hardware ecosystem: one centered on NVIDIA and its US-aligned partners, and a second centered on Chinese domestic alternatives increasingly positioned for export to countries outside the US sanctions net. The Global South is already becoming a contested market for both ecosystems.