China's Domestic AI Chip Surge Is a Policy Story, Not a Product One

When policy replaces competition as the primary driver of market share, the rules for forecasting change entirely.

🗄️ The Setup: A 41% Share Milestone Built in Three Years

Decoded:

China's AI chip market underwent a structural shift in 2025 that conventional market analysis missed: domestic vendors collectively shipped 1.65 million AI accelerator cards — 41% of China's total 4-million-unit market — according to IDC data reviewed by Reuters on April 1. Nvidia retained overall leadership, shipping 2.2 million cards for a 55% share. AMD captured a modest 4% with roughly 160,000 cards.

Huawei Technologies led all domestic vendors by a wide margin, shipping approximately 812,000 AI chips — roughly half of all Chinese-branded shipments. Alibaba's chip design unit T-Head ranked second at approximately 265,000 cards. Baidu's Kunlunxin and Cambricon each shipped around 116,000 cards. GPU startups Hygon, MetaX, and Iluvatar CoreX accounted for the remainder. China now has six domestic AI accelerator vendors with material production volumes — a manufacturing ecosystem that did not exist in this form four years ago.

The speed of this shift is the story. In 2021, before the first major U.S. export control round, Nvidia and AMD together held well over 90% of China's AI chip market. Four years later, domestic producers have crossed 40% — a faster share transfer than occurred in any other semiconductor category following U.S. restrictions. (Reuters/IDC, April 1, 2026)

🏛️ The Mechanism: Two Forces, Not One

China's 41% domestic share is driven by two distinct policy vectors operating simultaneously — and conflating them leads to wrong conclusions about where the market goes next.

The first force is subtraction. The October 2022 U.S. export control rules initially banned Nvidia's A100 and H100 training chips. Successive rounds restricted downgraded export variants — the A800 and H800 — in October 2023. The H20, designed specifically to meet export control thresholds on interconnect bandwidth, was banned in April 2025 after regulators determined China could cluster them to approximate restricted performance levels. Each ban removed the commercially viable Nvidia tier available to Chinese buyers, progressively compressing what Nvidia could sell inside China to nothing more than consumer and workstation GPUs.

The second force is substitution. China's central government launched a new AI infrastructure spending wave in 2025, with local governments accelerating intelligent computing centers under implicit directives to purchase domestic chips. These are not public mandates in most cases — they operate through procurement preferences embedded in government funding conditions, approved vendor lists for state-owned enterprise AI deployments, and guidance from Beijing's digital economy bureaus. The combination of government AI capital spending and buy-Chinese procurement routing created a structurally guaranteed demand pool that did not exist before 2023.

The critical investor implication: Nvidia's China retreat is not primarily a competitiveness failure. Huawei's Ascend chips do not outperform Nvidia's best products on standard AI benchmarks. The domestic vendors captured market share because the policy environment removed their primary competitor from the market and then channeled procurement toward what remained. That makes Nvidia's path back in China a diplomatic variable, not a product one.

📊 Investor Verdict

Why it matters: This IDC unit data quantifies what the impairment looks like in shipped volume: 2.2 million cards in a market that Nvidia would likely have supplied at 3.5–3.8 million units without restrictions, based on pre-2022 share trajectory. The forward risk is compounding. Reuters reported on March 27 that ByteDance and Alibaba are planning additional Huawei Ascend orders following successful testing — which would push domestic share past 41% in 2026 without any further U.S. policy action required. Huawei's March 31 annual report showed 22% revenue growth, partly driven by its chip and cloud infrastructure business — confirming Ascend is scaling toward hyperscaler-volume deployment.

Watch for: Two signals matter. First, any U.S.-China trade framework that creates a licensed export pathway for non-advanced Nvidia chips would offer a partial revenue recovery opportunity — though government buy-Chinese preferences would continue to apply independent of any product availability change. Second, quarterly disclosures from Alibaba, Baidu, and ByteDance on domestic chip procurement are the earliest visible indicators of whether domestic market share consolidates above 50% in 2026.

Stay decoded. See you tomorrow.

— The Get AI Decoded Team