A Bernstein estimate cited by The Economist shows China’s AI-chip market reorganizing around domestic suppliers — and the structural logic behind that shift matters more than the headline numbers.
A chart published by The Economist, drawing on Bernstein estimates and relayed by the South China Morning Post, offers the clearest single-frame picture yet of how US export controls have restructured China’s AI-chip market. By 2026, Huawei is estimated to hold roughly 50% of the market at approximately $12.1 billion in sales. Nvidia — which held around 39% as recently as 2025 — has fallen to roughly 8%, or about $2.0 billion. That is not a gradual decline; it is a near-total displacement in the span of a single year.
The remainder of the market is largely domestic. Cambricon accounts for an estimated $2.1 billion, Hygon/Sugon roughly $2.0 billion, with Alibaba’s in-house ASICs at ~$1.2 billion and Baidu’s at ~$0.7 billion. AMD, at an estimated $3.0 billion, is the largest remaining foreign supplier — larger than Nvidia’s restricted position. Taken together, domestic vendors reach approximately 56% of the market (up from ~46% in 2025), with Chinese-designed ASICs adding roughly another quarter. These are Bernstein estimates, not audited financials; treat them as directional signals, not precise measurements.
The trajectory is as significant as the snapshot. Cambricon is reportedly tripling output toward roughly half a million accelerators. Analysts project China will produce more domestic AI chips than its own demand requires by 2028, with domestic sales growing at an estimated 74% CAGR. That last figure, if it holds, converts a captive market into a potential export platform.
The Displacement Timeline
2025
Nvidia holds ~39% of China’s AI-chip market. Domestic vendors combined at ~46%. Export controls on H100/H800-class GPUs already in force; grey-channel and approved silicon still flowing.
Late 2025 – Early 2026
Huawei Ascend ramp accelerates. Cambricon triples production targets toward ~500,000 accelerators. Chinese labs — Alibaba, ByteDance, DeepSeek — increasingly rationed on Nvidia parts, accelerating procurement shift to domestic silicon.
2026 (Bernstein est.)
Huawei reaches ~50% (~$12.1bn). Nvidia falls to ~8% (~$2.0bn). Domestic vendors combined: ~56%. AMD surpasses Nvidia as largest foreign supplier at ~$3.0bn. Chinese ASICs (Alibaba + Baidu) add ~$1.9bn.
2028 (projected)
Analysts project China produces more domestic AI chips than domestic demand — at ~74% CAGR in domestic sales. Potential surplus becomes a potential export platform.
2026 China AI-Chip Market Share — Estimated Distribution
Huawei~50% / ~$12.1bn
AMD~$3.0bn
Cambricon~$2.1bn
Hygon / Sugon~$2.0bn
Nvidia~8% / ~$2.0bn
Alibaba ASIC~$1.2bn
Source: Bernstein estimates via The Economist / SCMP. Figures are estimates, not audited results. Baidu ASIC (~$0.7bn) not shown separately. Percentages reflect share of China’s AI-chip market specifically.
The key insight: Nvidia’s fall from ~39% to ~8% in a single year is not a competitive loss in the conventional sense — it is the export-control regime made visible as market share. Huawei did not out-engineer Nvidia in an open market; US policy restricted Nvidia’s best products from entering China and handed Huawei a captive customer base. Market share under those conditions is a policy outcome as much as a performance outcome. The capability gap and Nvidia’s CUDA software moat remain real. But captive markets have a way of closing capability gaps over time.
The Structural Read
The standard framing of US chip export controls is that they slow China’s AI development by denying access to leading-edge compute. That framing is not wrong, but the Bernstein data introduces a second-order effect that is underweighted in most policy analysis: the controls also accelerate the construction of a fully domestic, sanction-resistant compute stack — and they do so at a pace that market forces alone would not have achieved.
Three structural dynamics are now visible in the data.
First, the controls produced a domestic champion. Huawei’s ~50% share is not evidence that Huawei’s Ascend chips are performance-equivalent to Nvidia’s H100 or B200 series — they are not, on current benchmarks. What it does show is that a protected market with mandatory domestic procurement is a powerful industrial policy instrument. Nvidia’s restriction handed Huawei a revenue base and a volume ramp that it could not have acquired in an open competitive market. Import substitution, which the controls nominally aim to prevent, was accelerated by the controls themselves.
Second, this is the supply side of China’s sovereign AI stack. The hardware picture in this chart connects directly to several adjacent developments. DeepSeek’s reported ~$71 billion valuation and domestic compute raise signals where the demand for this silicon is going. The Huawei Ascend path — once treated as a contingency — is now the primary procurement channel for major Chinese labs. The H200 rationing that Alibaba, ByteDance, and DeepSeek faced was not a temporary friction; it was the forcing function that accelerated the shift to domestic silicon. And Chinese open-weight models are already capturing meaningful global token share on OpenRouter. The hardware and software layers of a sovereign stack are being assembled in parallel.
Third, the bifurcation may not stay contained. If the 2028 surplus projection holds — and it is a projection, not a certainty — China moves from import-substitution to potential export. The price-sensitive Global South markets already adopting low-cost Chinese AI models would be the natural early customers for lower-cost domestic Chinese silicon. That is not a near-term threat to Nvidia’s global dominance; Nvidia still leads by a wide margin outside China, and the CUDA software ecosystem is a genuine moat that hardware alone cannot quickly replicate. But it is the forward risk that the current market-share data foreshadows.
The Permission Layer — Applied
“Government regulation does not merely slow AI deployment — it actively shapes which companies win, which supply chains get built, and which ecosystems achieve scale. The export-control regime is not neutral policy friction; it is an industrial policy lever that simultaneously weakens Nvidia’s China position and strengthens Huawei’s. Understanding AI market structure in 2026 requires reading the policy layer first.”
Gennaro is the creator of FourWeekMBA, which reached about four million business people, comprising C-level executives, investors, analysts, product managers, and aspiring digital entrepreneurs in 2022 alone | He is also Director of Sales for a high-tech scaleup in the AI Industry | In 2012, Gennaro earned an International MBA with emphasis on Corporate Finance and Business Strategy.
Scroll to Top
Discover more from FourWeekMBA
Subscribe now to keep reading and get access to the full archive.
