AINEWS
Nvidia's China AI Chip Sales Stall as Huawei Takes the Lead
Bernstein projects Nvidia's share falling to about 8% this year, roughly matched by Huawei.
Bernstein projects Nvidia's share falling to about 8% this year, roughly matched by Huawei.
Nvidia's share of China's market for the artificial-intelligence chips that train and run large language models has collapsed from roughly 95% before U.S. export controls to around 40% today, and Bernstein, as reported by the Los Angeles Times, now projects it falling to about 8% by the end of this year, with Huawei roughly matched.
That is the doorway into a market reallocation that has happened unusually fast for an industry used to measuring change in quarters and process nodes. Two forces, working in parallel, did the work. Washington tightened the rules on which advanced chips Nvidia could ship to Chinese buyers. Beijing then turned that constraint into industrial policy, telling state-linked buyers and cloud customers that domestic alternatives, principally Huawei's Ascend line, were the preferred path. The first force blocked Nvidia from selling into a market it had effectively owned. The second made sure the demand that remained had somewhere local to land.
In late May, Nvidia's chief executive, Jensen Huang, traveled to Beijing for a closely watched visit that coincided with a Trump-Xi summit. The trip produced a single piece of enduring color: a public appearance at a Beijing noodle shop where Huang ate zhajiangmian, the wheat-noodle dish with soybean-paste sauce that is a fixture of the capital's casual dining. The optics, a U.S. chip executive eating noodles in the middle of a U.S.-China technology dispute, drew the kind of attention that handshakes usually do not.
The substance was more pointed. In remarks to Chinese reporters, Huang put the share collapse in his own words: "Before the export control banned Nvidia out of China, we had about 95% market share, and so we were competing just fine." He paired that concession with a national-security caveat that matters for how the policy debate reads: the United States, he said, has to balance security against the export growth that funds its technology lead, and the current approach is short on that balance.
Huang also called Chinese competitors, Huawei foremost among them, "giants," language that even sympathetic analysts read as an acknowledgment that the contest inside China is no longer a mismatch.
The numbers behind his framing are now coming into public view through analyst projections carried in the LA Times's reporting. Bernstein's latest data tracks Nvidia's share in China's advanced AI chip market sliding from roughly 95% to around 40% today, with a year-end estimate near 8%, and Huawei now running close to even. The collapse is steeper than the U.S. policy debate usually acknowledges, and it has happened while China's demand for AI compute has, if anything, accelerated.
That demand has a name on the model side, and it is not a U.S. one. DeepSeek's recent models, which deliver competitive performance at substantially lower training cost, have become the reference point for Chinese cloud buyers choosing what hardware to standardize on. Technical comparisons of Huawei's Ascend 910D against Nvidia's H100 frame the older Huawei part as a credible domestic alternative rather than a generation behind. A newer Ascend 950PR, targeted at the ~1.56 PFLOP class, is now being positioned as the chip that resets the comparison, though the figure comes from secondary technical-comparison pages and has not yet been confirmed by a Huawei datasheet or independent benchmark.
The policy backdrop deserves its own paragraph because it is the part of this story that is moving fastest. The Trump administration's export-control regime initially blocked sales of Nvidia's H200 chip to China. After negotiations tied to the May summit, the administration agreed to permit them, a partial reprieve for Nvidia. Beijing's response has been to harden, not soften, its preference for domestic procurement. State guidance to local cloud and telecom buyers has effectively narrowed the addressable market for any foreign chip, including the newly permitted H200. Nvidia's relief on the U.S. side and its losses on the Chinese procurement side are happening at the same time.
What to watch next: whether the Trump-Xi understanding translates into a durable framework for H200-class shipments or stays a one-off concession; whether Huawei's 950PR ships at volume on the timeline its promoters are projecting; and whether Bernstein's year-end 8% estimate holds or proves optimistic about how quickly Chinese buyers can finish the substitution.
The verdict-shaped language that this contest invites, that "China has won the AI chip war," does not survive contact with the data. What is happening is more specific, and in some ways more uncomfortable for U.S. policy: a major U.S. firm was denied access to a market it had owned, and the resulting vacuum has been filled at speed by a state-backed domestic champion. The Huang concession is news because it is the first time the chief executive of the world's most valuable chip company has put the mechanism in plain words.