TL;DR
- Market Share: Chinese domestic chipmakers captured 41 percent of China’s AI accelerator market in 2025, shipping 1.65 million cards, according to IDC data.
- Huawei’s Lead: Huawei dominated domestic shipments with roughly 812,000 chips, backed by doubled yield rates and government-directed SMIC production priority.
- Export Controls: Escalating US chip export restrictions have driven Chinese data center operators toward domestic alternatives, with Washington now considering universal approval requirements for all semiconductor exports.
- Nvidia’s Position: Nvidia’s China share stands at 55 percent but could fall to as low as 8 percent under sustained restrictions, according to Bernstein analyst projections.
Nvidia’s grip on China’s AI chip market is slipping. Domestic vendors shipped 1.65 million accelerator cards in 2025, capturing 41 percent of the market and closing in on Nvidia’s 55 percent share, according to IDC data reported by Reuters.
About 4 million AI accelerator cards were shipped in China last year, with Nvidia moving roughly 2.2 million units. Just a few years ago, Nvidia held near-total dominance over the Chinese AI accelerator market. Bernstein analysts estimated in January 2026 that Huawei and Nvidia were roughly tied at about 40 percent each. IDC now confirms domestic vendors have collectively surpassed that estimate, reaching 41 percent when smaller players beyond Huawei are counted.
For Nvidia, a company that once counted China as the source of roughly one-fifth of its data center revenue, the erosion carries strategic consequences extending well beyond a single market. China remains one of the largest AI hardware deployment regions in the world, and a shrinking foothold there compounds pressure from export restrictions that have already curtailed which products Nvidia can legally ship. Analysts tracking the market now debate not whether domestic vendors will gain share, but how quickly the transition will accelerate.
US export controls launched a market transformation that was barely visible three years ago. Restrictions initiated under the first Trump administration and extended under Biden progressively closed off China’s access to Nvidia’s most advanced chips, then expanded to cover the downgraded alternatives the company designed specifically for the Chinese market. That urgency now shows up in shipment data. Domestic chipmakers capturing more than four in ten accelerator cards sold would have been unthinkable at the outset of those restrictions.
China’s localization drive is gaining further impetus from allied chip export controls, which have closed potential workarounds through third countries and broadened pressure on Chinese buyers to seek domestic sources.
Huawei Leads a Growing Domestic Pack
Huawei dominates the domestic field with about 812,000 chips shipped in 2025, accounting for roughly half of all domestic shipments. Alibaba’s chip unit T-Head came second with 265,000 cards, while Baidu Kunlunxin and Cambricon were tied at 116,000 units each. AMD held just 4 percent of the overall market.
Huawei alone shipped more than three times as many cards as T-Head, the largest domestic vendor after it, underscoring how concentrated domestic production remains even as the overall field expands. The three-to-one gap illustrates the challenge Beijing faces in building a diversified chip supply chain: government policy is creating a national champion faster than it is creating a competitive domestic ecosystem.
Huawei’s position extends beyond raw shipment numbers. The company has doubled its AI chip yield to nearly 40 percent, up from 20 percent a year ago, easing production bottlenecks that had previously constrained output. Huawei achieves this on older lithography processes at SMIC, China’s leading chip foundry. Analyst Paul Triolo noted that SMIC is prioritizing Huawei production, probably under direction from industrial planning authorities, giving Huawei a structural manufacturing advantage that smaller domestic firms like Cambricon and Baidu’s Kunlunxin cannot easily replicate.
Bernstein has outlined a three-year plan for Huawei to overtake Nvidia entirely in China, a projection that looked ambitious when published in January but appears more credible against the backdrop of IDC’s full-year 2025 data. Huawei’s Ascend 910C chips now power approximately 50 percent of domestic data centers in China, reflecting deliberate government-backed deployment rather than organic commercial adoption alone. The Ascend 910C delivers approximately 60 percent of Nvidia’s H100 inference performance, a gap that matters for advanced training workloads but is increasingly acceptable for the inference tasks dominating commercial deployment. For the majority of enterprise AI workloads, which involve running existing models rather than training new ones, a 40 percent performance shortfall is a manageable trade-off against the practical unavailability of Nvidia hardware.
Software ecosystem development is accelerating alongside the hardware gains. Chinese AI firm Zhipu trained a major model on Huawei chips, developing its GLM-5 entirely on Huawei Ascend hardware using the MindSpore framework, without relying on any US-manufactured semiconductors. Shenzhen has launched a 10,000-card AI cluster built entirely on Huawei hardware, demonstrating that government-backed deployment is moving from pilot projects to production scale. Until recently, Huawei’s Ascend chips were widely dismissed as uncompetitive by Western analysts; the shipment data and ecosystem adoption tell a different story.
Export Controls Fuel the Shift
Market realignment traces directly to tightened US export controls and Beijing’s complementary push for domestic chip adoption. Controls initiated under the first Trump administration and extended under Biden have progressively restricted what Nvidia can sell in China, first targeting advanced chips, then expanding to cover the downgraded H20 that Nvidia designed specifically to stay within prior export thresholds. US restrictions have also broadened to cover allied nations’ semiconductor equipment exports, tightening the supply chain at multiple points and narrowing options for Chinese buyers seeking foreign hardware.
Washington is now reportedly considering sweeping new chip export controls requiring federal approval for every semiconductor export sale. The proposed centralized approval process would shift from targeting specific chip performance thresholds to requiring government sign-off on all semiconductor exports, affecting both Nvidia and AMD across their entire product lines. A separate bill to ban the sale of key AI chipmaking equipment to China has also been introduced in the US House, signaling bipartisan momentum behind tighter restrictions.
Beijing actively pushes data centers to use domestic chips through energy subsidies, procurement mandates, and regulatory pressure. State-backed energy subsidies lower operating costs for data centers that commit to domestic hardware, effectively making the economics of choosing Huawei over Nvidia more attractive even where performance gaps exist. Chinese data center operators face both a ceiling on how many Nvidia chips they can obtain and a floor on how many domestic chips they must deploy, a two-pronged policy structure explaining why market substitution has been faster than pure technology competition would predict.
Building on those constraints, Nvidia has responded by halting China-bound H200 output and redirecting TSMC capacity to its upcoming Vera Rubin platform, effectively deprioritizing the Chinese market in its production planning. Analyst Randy Abrams noted in January that limiting Nvidia’s access to China would fuel the growth of domestic alternatives, and the IDC data now validates that assessment. Each new round of tighter controls appears to push more Chinese buyers toward Huawei and other local vendors, accelerating the substitution cycle.
Bernstein’s longer-range projection puts the stakes in sharp relief: under sustained export restrictions, Nvidia’s China share could fall to as low as 8 percent while Huawei’s could climb to approximately 70 percent. With US and allied governments tightening controls in parallel and Beijing subsidizing domestic adoption, the trajectory favors continued substitution in one of the world’s largest AI hardware markets.
DeepSeek’s emergence in 2025 had already challenged assumptions about China’s ability to innovate under chip restrictions, demonstrating that algorithmic ingenuity could compensate for hardware constraints in ways Western observers underestimated. That software breakthrough and the hardware market share data now tell a consistent story: China is making rapid, broad-based progress across the AI stack, not just at the application layer. IDC’s shipment figures show the hardware supply chain following a similar trajectory of rapid catch-up. At the current pace, 41 percent may prove not a ceiling but a waypoint on the path toward a much deeper shift in global AI hardware dominance.
