Huawei Claims Ascend Topped Nvidia In China As US Export Curbs Bite

Nvidia's H200 shipments to China, the first ever approved under the Trump administration, amounted to less than 1 percent of the company's data center revenues. Executives said so on Nvidia's most recent quarterly earnings call, and The Register reported the figure on 30 September. It is the clearest sign yet that the export control regime built in Washington has begun to reshape the Chinese accelerator market from the inside.

The Register's report centers on a Q&A-style release from Huawei this week. In it, rotating chairman Eric Xu claimed the company's Ascend line of neural processing units had exceeded Nvidia in Chinese market share. "It's pretty hard to collect data about the market share of Nvidia in China, but based on the data we have collected, Ascend has surpassed Nvidia," Xu said, according to the report. No third-party market share dataset validating the claim appears in the dossier. Nvidia has not publicly confirmed or disputed it.

A software stack built for Ascend

The claim lands one day after DeepSeek, the Hangzhou AI developer, open-sourced a suite of six software modules tailored for Huawei's Ascend chips. The South China Morning Post reported on 30 September that the release mirrors DeepSeek's earlier open-source tools for Nvidia hardware. Its stated purpose is to build an "independent and controllable" software ecosystem for AI processors.

Among the modules is an Ascend-compatible version of TileLang, a programming language for high-performance kernels. TileLang lists Nvidia as its primary back end but now officially supports Huawei's Ascend 950 accelerators, according to an update on the project's GitHub page cited by the SCMP. That matters because chip market share is not decided by silicon alone. Developers write kernels, tune schedulers and maintain toolchains. If those tools assume a different vendor's architecture, switching costs stay high regardless of how many cards a buyer can procure. DeepSeek's release, six modules in total, is an attempt to lower that cost for Ascend. It is also a hedge, because the same export controls that pushed Chinese buyers toward Huawei can, in principle, be tightened again at any moment.

Xu framed the supply argument directly. "Even though our chips may be less advanced, at least their supply is assured, so that you don't have to worry about chip supply day in and day out," he said, according to The Register.

"For the Chinese government, for the domestic industry, and for Huawei, the path forward is undoubtedly to push for full self-sufficiency in terms of chips and the entire semiconductor value chain."

How Washington got here

The policy path has been anything but linear, and the dossier's timeline is worth laying out because each reversal changed buyer behavior. In April 2025 the US Commerce Department imposed new licensing requirements on Nvidia's and AMD's China-spec accelerators. That summer the department reversed course, and the administration later reached an arrangement to cut the US government in on a percentage of certain China-bound chip sales. By late last year, the administration announced Nvidia would be allowed to sell its more potent H200-series parts to approved Chinese customers for the first time ever.

By then, the damage to Nvidia's position was already done, The Register reported. The initial blockade of H20s earlier in 2025 set off alarm bells in Beijing. Government officials began pressuring data center operators to transition away from foreign accelerators in favor of homegrown alternatives. The H200 approval arrived after that shift had started. Nvidia's own earnings disclosure, that H200 China shipments came in below 1 percent of data center revenue, suggests the market did not snap back.

Beijing's own signals have been mixed. Headlines in the dossier note that Beijing signaled approval for Nvidia's RTX Pro 5500, a chip expected to dodge US curbs. Separate reports on 29 September said Nvidia and AMD were lobbying the White House to halt potential chip export restrictions to China tied to a defense bill. Those headlines sit outside the full-text sources and cannot be cited for facts here, but they indicate the lobbying fight is live.

Scale is Huawei's compensating strategy. The Register reported that Huawei is currently deploying a 256,000-card Atlas 950 SuperCluster, and that its newer architecture is designed to support training and inference for multi-trillion-parameter models and to scale to as many as one million NPUs. Xu also downplayed the idea of a near-term international push. "We don't have enough capacity to satisfy the demand in China. We don't have plans to expand the international market in a fully-fledged way," he said, according to the report.

The design layer moves too

Two other deals this week point at the layer above the fab. On 30 September, OpenAI and Synopsys said they had signed a multi-year strategic partnership to build a specialized AI model for chip design called GPT-Synopsys, according to The Decoder. Synopsys makes electronic design automation tools, the software engineers use to design chips.

The stated goal is a model that can "reason about chip design and verification, and to directly operate Synopsys' tools," with engineers delegating objectives and reviewing output. Customer data will not be used for training and will be stored encrypted, both companies said. Early tests with semiconductor customers are already underway.

Separately, EE Times reported on 30 September that Emergence AI is moving its neuroformal AI technology into active deployments with fabless semiconductor companies, and that integrated device manufacturers among its customers are asking to extend the work into the fab. Co-founder and executive chairman Satya Nitta told EE Times that demand exceeds supply and every fab is running at capacity, so the lever is yield rather than wafer volume. "You cannot make chips any faster, but what you can do is definitely get more chips per wafer yielding with AI," he said. Nitta described agents that spot the same failure pattern across roughly 30 percent of 1,500 products and propose a block redesign, and said the company pairs LLM-style generation with symbolic verification so answers are "provably correct."

Those are commercial bets, not policy. But they matter to the export control debate because they attack the constraint from a different direction. If design cycles shorten and yields improve, the effective capacity of whichever fabs are available goes up, and the relative cost of routing around restricted suppliers falls.

Voluntary controls in Washington

Meanwhile the US is trying to govern AI safety without new law. CBS News reported on 29 September that President Trump and the leaders of the largest US AI companies signed a set of voluntary safety standards after a White House lunch, with attendees including Elon Musk, Mark Zuckerberg, Dario Amodei, Jensen Huang, Greg Brockman and Sundar Pichai.

The one-page accord, posted by Trump to Truth Social, commits the companies to "four layers of controls and audits," including internal evaluations, audits by an external firm and reviews by each company's board, and to meeting regularly to set standards. Asked whether the deal was binding, Trump said: "I think it's morally binding." He also said the group discussed forming a 10-person committee to "watch over the enterprise," and that he plans to name an "AI czar" within three or four days. On the same day he signed an executive order directing the federal government to use the term "Super Intelligence" instead of artificial intelligence. Huang argued there was "no conflict between innovation, technology and safety."

The accord is a statement of principles, not a rulebook, and its own text acknowledges as much: "Over time, it may make sense to codify these steps into laws or regulations." For chip exporters, the near-term question is narrower. The controls that reshaped the Chinese market were unilateral US actions on hardware sales. They remain the instrument with the most direct effect on Huawei's and Nvidia's relative positions.

Europe's parallel fight

Outside the chip file, the EU is advancing its own contested tech rules. Reclaim The Net reported on 30 September that the Council is pushing a revised chat control regulation that would let police order internet companies to scan private communications in "parts of a service" for a period of time, rather than requiring a warrant targeting a specific suspect. The detail comes from a leaked Council presidency note dated 18 September, prepared for a trilogue on 29 September, with Commissioner for Internal Affairs Magnus Brunner representing the Commission.

Former Pirate Party MEP Patrick Breyer called the approach "mass surveillance by another name," and said scanning "parts of a service" would in practice cover "every user in the EU," according to Reclaim The Net. The same report notes the Council's own legal service has previously said scanning an entire service or parts of it is "highly probable" to be found "general and indiscriminate."

The two files are separate, but they share a pattern: governments reaching for broad technical mandates, and industry and civil society contesting the scope.

On the ground in China, the AI buildout sits against a weaker consumer economy. David Cheng, an investor who traveled to Beijing and Shanghai, wrote on his Earned Intuition newsletter on 30 September that youth unemployment stood at 17.9 percent as of July and that July retail sales rose 0.6 percent year on year against an expected 1.5 percent. His framing is that AI is "the only engine of the Chinese economy that is working," propped up by state-subsidized electricity and fast-tracked IPOs. That context does not settle the market share question, but it explains why Beijing keeps pushing self-sufficiency regardless of what Washington permits next.