Nvidia’s grip on China’s AI-chip market is slipping faster than most analysts expected. A Reuters report published October 13, 2026 cites new IDC data showing Nvidia held just 34% of China’s data-center AI-chip market in the first half of 2026, down sharply from the 55% share it commanded for all of 2025. Huawei, meanwhile, has climbed to 28% of the same market, more than doubling its unit shipments year over year. The numbers mark one of the clearest signs yet that Beijing’s push for chip self-sufficiency is translating into real market movement, not just policy rhetoric.
The shift matters well beyond China’s borders. Nvidia built its trillion-dollar valuation largely on the assumption that its chips would remain the default choice for AI infrastructure everywhere demand exists. China was never going to be a simple market for Nvidia given export restrictions, but the speed of Huawei’s advance suggests a structural change in how the world’s second-largest economy builds out AI compute. Here is what the new data shows, what remains unverified, and what it means for the broader AI hardware race heading into 2027.
The IDC Numbers Behind the Headline
According to the IDC report cited by Reuters, Nvidia shipped 678,000 AI accelerators in China between January and June 2026, a 44% drop from the same period a year earlier. Huawei shipped 572,000 AI accelerators over the same six months, up from more than 300,000 units a year prior. That single comparison, a 44% decline for Nvidia against roughly an 80% jump for Huawei, captures the pace of the reversal better than any single market-share percentage can.
IDC’s data also shows the broader domestic ecosystem consolidating fast. Huawei, T-Head (Alibaba’s chip-design unit), Kunlunxin (Baidu’s AI silicon arm), and Cambricon together accounted for 95% of China’s non-GPU AI-chip segment in the first half of 2026. That is a near-total lock on the category of accelerators that sit outside the traditional GPU architecture Nvidia pioneered, and it signals that Chinese cloud providers and enterprises are increasingly comfortable building on homegrown silicon rather than waiting for GPU supply to normalize.
| Vendor | H1 2026 Units Shipped (China) | Year-over-Year Change | Market Share |
|---|---|---|---|
| Nvidia | 678,000 | -44% | 34% |
| Huawei | 572,000 | Up from 300,000+ | 28% |
| T-Head, Kunlunxin, Cambricon (combined) | Not separately disclosed | Not disclosed | Part of 95% non-GPU segment share held by domestic vendors |
IDC projects total AI-accelerator shipments in China will grow 19% in 2026, even as GPU shipments specifically are expected to fall 9%. That divergence is the real story underneath the headline numbers. The overall pie is growing, but a shrinking slice of it is going to traditional GPU architecture, and an even smaller slice of that GPU slice is going to Nvidia.
How Fast Nvidia Fell: 2025 vs. the First Half of 2026
Context from the full 2025 calendar year makes the drop look even steeper. In 2025, Chinese GPU and AI-chip makers collectively shipped approximately 1.65 million AI-accelerator cards domestically, representing 41% of the market, according to an IDC report reviewed by Reuters. Nvidia, by contrast, shipped roughly 2.2 million cards in China that year and held a 55% share. Huawei alone shipped approximately 812,000 AI chips in 2025, about half of all domestically branded shipments, while T-Head shipped roughly 265,000 cards.
Line those two snapshots up and the trajectory is unmistakable. Nvidia went from 55% of the market across all of 2025 to 34% in just the first six months of 2026, a drop of 21 percentage points in roughly half a year. Huawei’s unit shipments alone, 572,000 in six months versus 812,000 for the entirety of 2025, put it on pace to roughly match or exceed its full prior-year total before 2026 is even over.
| Period | Nvidia Share | Nvidia Units | Domestic Chinese Vendors (Combined Share) |
|---|---|---|---|
| Full Year 2025 | 55% | ~2.2 million | 41% |
| H1 2026 | 34% | 678,000 | Huawei alone at 28%; non-GPU domestic vendors at 95% of that sub-segment |
It’s worth being precise about what these two data sets measure. The 2025 figures cover the full calendar year and include both GPU and non-GPU accelerators across all domestic vendors. The H1 2026 figures isolate Nvidia and Huawei specifically within the same category. The comparison is directionally consistent, both show Nvidia losing ground and Chinese vendors gaining it, even if the exact percentage-point math isn’t a perfectly clean apples-to-apples read across the two reporting periods.
Huawei’s Own Claim: Ascend Has “Surpassed” Nvidia
Huawei hasn’t been shy about framing the IDC numbers as validation of a bigger claim. Eric Xu, Huawei’s rotating chairman, said that collecting precise data on Nvidia’s China market share is difficult, but based on what Huawei has gathered, “Ascend has surpassed Nvidia,” according to a report from The Register. Xu went further, stating that Huawei’s Ascend market share “should be bigger than Nvidia’s.” Reuters, which also covered Xu’s comments, reported that Huawei did not provide data supporting that broader claim.
That gap between what IDC’s numbers show and what Huawei’s chairman is saying publicly is worth sitting with. The IDC data puts Huawei at 28% and Nvidia at 34% for the first half of 2026, meaning Nvidia still leads on paper. Xu’s framing suggests Ascend has already overtaken Nvidia outright. Both statements can be technically defensible depending on how “market” is defined, whether by unit count, by revenue, by data-center deployments versus edge devices, or by a narrower slice of the AI-accelerator category. TrendForce reported on Huawei’s broader claim alongside separate comments the company made about backing NPO over CPO for optical interconnects, suggesting Huawei is pushing its market-share narrative on multiple fronts at once.
The Unconfirmed Bernstein Estimates
Separate from the IDC figures, some reports have cited Bernstein Research estimates placing Huawei’s Ascend share of China’s AI-accelerator market at roughly 50%, with Nvidia falling to around 8%. Those numbers are considerably more dramatic than the IDC figures and are not independently confirmed. A piece from Startup Fortune and coverage from other outlets have repeated versions of this claim, but the underlying methodology behind the 50%/8% split hasn’t been made public in the way IDC’s survey-based figures have.
This is the single most important distinction to hold onto when reading coverage of this story: the claim that Huawei Ascend has definitively overtaken Nvidia in China is not independently confirmed by the available data. Huawei itself has acknowledged that Nvidia’s exact market share is hard to pin down. Readers should treat any specific number north of Huawei’s IDC-reported 28% with appropriate skepticism until a methodology-transparent source confirms it.
Why the Gap Is Closing So Fast
The mechanics behind this shift are not mysterious even if the exact magnitude is contested. US export controls have restricted which Nvidia chips can legally ship to Chinese customers for several years now, forcing Nvidia to sell down-specced variants in the market or, at various points, none at all. That policy backdrop created the opening Huawei needed, and Huawei’s Ascend line has had years to mature inside a protected home market with heavy state backing.
Chinese cloud operators and AI labs also have strong incentive to diversify away from a single foreign supplier whose access to their market can change with US policy shifts. Our earlier coverage of DeepSeek shipping Ascend-compatible developer tools to rival CUDA showed this dynamic playing out at the software layer months before the latest IDC numbers confirmed it at the hardware level. When a major Chinese AI lab builds tooling specifically to make Ascend chips easier to program, that is a leading indicator of exactly the share shift IDC is now reporting.
Nvidia’s Response: Open Source as a Counterweight
Nvidia has not been passive while its China share erodes. The company has leaned into open-source AI model investment as a way to keep developers anchored to its software stack even where its chips can’t ship freely. Tech Insider reported on Nvidia’s $20 billion open-source push aimed at countering China’s lead in widely-used open models, a strategy that works around the hardware restriction by keeping the broader CUDA ecosystem relevant to developers who might otherwise migrate fully to Ascend-based tooling.
Nvidia has also been active on the acquisition front, reportedly in talks for a stake in Reflection AI. Tech Insider covered both the $25 billion stake talks and the earlier reported $800 million acquihire interest in the company. Reflection AI itself has positioned its open releases as a direct answer to China’s open-model momentum, something we detailed in our report on Reflection AI’s bet to take on China’s open models. None of these moves directly reverse the unit-shipment numbers IDC reported, but they represent Nvidia’s attempt to defend the software and developer-ecosystem layer even as its hardware share in China declines.
China’s Side of the Equation: Will Beijing Loosen Up?
There’s an added wrinkle that complicates the simple “Huawei is winning” narrative. A report from The Chosun Daily indicates China is weighing whether to allow more Nvidia GPU imports to accelerate domestic AI development, suggesting Chinese policymakers and companies are not uniformly committed to a full substitution strategy. Even as Huawei and the broader domestic chip ecosystem gain share, there appears to be recognition inside China that cutting off Nvidia entirely could slow AI progress at a moment when the country is racing to keep pace with US frontier labs.
That tension, between wanting chip sovereignty and wanting the fastest possible AI compute, is likely to keep playing out through 2026 and into 2027. If Beijing does relax import restrictions even modestly, Nvidia’s unit numbers could stabilize or partially recover without any change to the underlying export-control posture from Washington.
The Non-GPU Chip Race Inside China
While Nvidia versus Huawei gets most of the attention, the IDC data points to something arguably more significant for China’s long-term AI strategy: near-total domestic control of the non-GPU accelerator category. Huawei, T-Head, Kunlunxin, and Cambricon holding 95% of that segment means Chinese AI infrastructure is increasingly built on architectures Nvidia doesn’t compete in at all, not just chips that substitute for Nvidia’s GPUs on a one-to-one basis.
| Company | Chip Unit / Parent | Segment Role |
|---|---|---|
| Huawei | Ascend | Largest non-GPU vendor by shipment volume in China |
| T-Head | Alibaba chip-design unit | Shipped ~265,000 cards in 2025 |
| Kunlunxin | Baidu AI silicon arm | Part of 95% non-GPU domestic combined share, H1 2026 |
| Cambricon | Independent chip designer | Part of 95% non-GPU domestic combined share, H1 2026 |
This matters because it diversifies China’s AI compute supply chain across four separate domestic companies rather than concentrating all substitution risk in a single vendor. If any one of these four faces its own supply constraints, whether from advanced lithography access or packaging capacity, the other three provide some cushion. That kind of structural resilience is harder to build than market share alone and suggests China’s chip strategy is more deliberate than a simple nationalist preference for homegrown silicon.
Market Impact: What This Means for Nvidia’s Business
China has historically been a meaningful but not dominant slice of Nvidia’s total data-center revenue, with most of Nvidia’s AI chip demand coming from US hyperscalers and enterprises. The China share decline matters less for Nvidia’s near-term earnings than it does for the long-term precedent it sets. A report from U.S. News, covering an earlier stage of this same trend back in June 2026, framed Nvidia’s China sales as stalling specifically because local chipmakers were taking the lead, a framing the new IDC data now confirms with harder numbers five months later.
The bigger risk for Nvidia is reputational and strategic rather than purely financial in the short term. If Huawei’s Ascend line proves it can scale production and software maturity inside China’s protected market, that same playbook becomes a template other countries facing US export restrictions could look to replicate. Investors watching Nvidia’s stock have generally shrugged off China-specific headlines so far given the company’s dominant position in the US and allied markets, but a sustained, confirmed loss of China share removes one more growth lever from a company that has leaned heavily on expanding total addressable market as its core investment story.
Competitive Landscape: Nvidia, AMD, and the Rubin Pipeline Elsewhere
It’s useful to put the China numbers in context against what Nvidia is doing in markets where it faces no export restrictions. Tech Insider reported on AM Intelligence’s order of 20,000 Nvidia Rubin GPUs for its India buildout, a deal that shows Nvidia’s next-generation Rubin architecture moving at full scale in markets outside China’s restricted zone. That contrast, strong Rubin demand in India against a shrinking GPU share in China, underscores that Nvidia’s global position isn’t uniformly weakening. It’s specifically concentrated in the one major market where policy, not product competitiveness, is the primary constraint.
Nvidia’s broader chip strategy has also shown a pattern of managing product transitions aggressively when market conditions shift. Our coverage of the RTX 5090’s production exit echoing Nvidia’s eight-year chip playbook showed how the company reallocates fab capacity and shifts product lineups in response to demand signals. Whether a similar recalibration happens for China-specific chip variants remains to be seen, but Nvidia has the organizational muscle memory to pivot quickly if the China numbers continue sliding.
Historical Context: How We Got Here
Nvidia’s China dominance was never guaranteed to last indefinitely once Washington began tightening export rules on advanced AI chips starting in 2022 and 2023. Each successive round of restrictions forced Nvidia to design China-specific chip variants with reduced capability, a workaround that kept some revenue flowing but also gave Chinese buyers a clear signal that relying on Nvidia long-term carried political risk outside their control. Huawei, sanctioned more broadly and for longer than Nvidia has faced restrictions, had already built years of experience operating under constrained access to advanced manufacturing before the AI chip boom even started, making it arguably the best-positioned domestic company to absorb displaced demand once Chinese buyers started actively seeking alternatives.
The 2025 full-year figures, Nvidia at 55% and domestic vendors at a combined 41%, already showed this shift underway well before the newest H1 2026 numbers. What’s changed is the pace. A 21-percentage-point drop in Nvidia’s share within roughly six months is a much sharper move than the gradual multi-year erosion many analysts had modeled.
Predictions: Where This Goes Through 2027
- Nvidia’s China AI-chip share likely continues declining through the second half of 2026, though probably at a slower rate than the H1 drop given how much ground has already shifted.
- Huawei’s Ascend shipment volume for full-year 2026 will likely exceed its 2025 total of roughly 812,000 units, based on the 572,000 units already shipped in just the first six months.
- Expect continued disagreement between IDC-style survey data and company-sourced claims like Huawei’s “surpassed Nvidia” framing, with the gap between the two narratives unlikely to fully close without a neutral, methodology-transparent audit.
- If China eases Nvidia GPU import restrictions as suggested by reports like the one from The Chosun Daily, Nvidia’s unit numbers could partially stabilize in 2027 without any change to US export policy.
- The non-GPU domestic chip segment, where Huawei, T-Head, Kunlunxin, and Cambricon already hold 95% combined share, will likely remain almost entirely closed to foreign vendors regardless of what happens with GPU-specific import policy.
What Enterprise AI Buyers Outside China Should Watch
For AI infrastructure buyers outside China, the direct impact of this shift is limited in the near term since export restrictions already prevent most cross-border chip comparisons from mattering operationally. The more relevant signal is what Huawei’s rapid domestic scaling says about the maturity of alternative AI accelerator architectures generally. If Ascend chips can hit 572,000 units shipped in six months inside a market with strong state support, that’s evidence non-Nvidia AI accelerator architectures can reach meaningful production scale faster than skeptics previously assumed, a data point relevant to any company evaluating AMD, custom silicon, or other Nvidia alternatives for reasons unrelated to China policy at all.
It’s also worth watching whether AI model training results coming out of Chinese labs running primarily on Ascend or other domestic silicon start matching Nvidia-trained models on benchmark performance. If domestic Chinese AI labs demonstrate they can train frontier-competitive models on 95%-domestic-silicon infrastructure, that would be a stronger signal of long-term technological parity than shipment numbers alone can provide.
The Verification Gap: Why Numbers Keep Conflicting
Readers trying to track this story across multiple outlets will keep running into inconsistent numbers, and it’s worth understanding why. IDC’s figures come from a standard market-research methodology based on shipment tracking and vendor surveys, the same approach used across most hardware market-share reporting. Huawei’s own claims, including Eric Xu’s comments, are company statements without published underlying data. The Bernstein-style 50%/8% estimates sit somewhere in between, produced by financial analysts but without full methodology disclosed publicly in the reporting that has circulated.
None of these sources are necessarily wrong, they may simply be measuring different things: total units versus data-center-specific deployments, revenue versus volume, or different time windows. Until a single standardized measurement becomes the accepted industry benchmark, expect the “who’s really winning in China” debate to keep generating conflicting headlines even as the broader direction, Nvidia losing share, domestic vendors gaining it, remains consistent across every version of the data.
FAQ
What is Nvidia’s current AI-chip market share in China?
According to an IDC report cited by Reuters, Nvidia held 34% of China’s data-center AI-chip market in the first half of 2026, down from 55% for the full 2025 calendar year.
What is Huawei’s AI-chip market share in China?
IDC data puts Huawei at 28% of China’s AI-chip market in the first half of 2026, up from a smaller base in 2025 when Huawei shipped approximately 812,000 AI chips for the full year.
Has Huawei’s Ascend chip actually overtaken Nvidia in China?
That claim is not independently confirmed. Huawei chairman Eric Xu said Ascend “has surpassed Nvidia,” but Huawei provided no supporting data, and the confirmed IDC figures still show Nvidia ahead at 34% versus Huawei’s 28% for the first half of 2026.
Why is Nvidia losing AI-chip market share in China?
US export controls on advanced AI chips have restricted which Nvidia products can legally ship to China, while domestic vendors like Huawei have used the opening to scale production and software maturity within a protected home market.
Which other companies make AI chips in China besides Huawei?
T-Head (Alibaba’s chip-design unit), Kunlunxin (Baidu’s AI silicon arm), and Cambricon are the other major domestic players. Together with Huawei, these four companies held 95% of China’s non-GPU AI-chip segment in the first half of 2026, according to IDC.
Is China’s overall AI-chip market growing or shrinking?
IDC expects total AI-accelerator shipments in China to grow 19% in 2026, even though GPU-specific shipments are expected to fall 9% over the same period, reflecting a shift toward non-GPU domestic architectures.
Could Nvidia’s China market share recover?
It’s possible. A report from The Chosun Daily indicates China is weighing whether to allow more Nvidia GPU imports to accelerate domestic AI development, which could partially stabilize Nvidia’s unit numbers if adopted.
Where do the Bernstein estimates of Huawei holding 50% share come from?
Those figures are attributed to Bernstein Research and are not confirmed by the IDC data referenced in Reuters’ reporting. They should be treated as a separate, unverified estimate rather than an official market-share figure.