Huawei says Ascend AI chip sales in China have overtaken Nvidia

Huawei's rotating chairman Eric Xu claimed in a Q&A-style release this week that the company's Ascend line of neural processing units had exceeded Nvidia in Chinese market share, according to The Register, which reported the remarks on 30 September. "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.

The claim is hard to verify independently, and Xu gave no figures. Nvidia has not published a China-only share number. During its Q2 earnings call, executives said shipments of H200 accelerators to the region amounted to less than 1 percent of datacenter revenues, The Register reported.

DeepSeek releases tools for Ascend

The Huawei claim landed two days after DeepSeek open-sourced a suite of six software modules tailored for Huawei's Ascend chips, according to the South China Morning Post on 30 September. Among the releases is an Ascend-compatible version of TileLang, a programming language for high-performance kernels. The project's GitHub page now lists official support for Huawei's Ascend 950 accelerators, with native code generation, automatic scheduling and synchronisation, the SCMP reported. DeepSeek said the aim is an "independent and controllable" software ecosystem for GPUs.

That is the software half of a stack that US export controls pushed Beijing to build. After Washington imposed licensing requirements on Nvidia's and AMD's China-spec accelerators in April 2025, the Commerce Department reversed course that summer and 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 said Nvidia could sell more potent H200-series parts to approved Chinese customers for the first time.

By then, Chinese officials were already pressuring datacenter operators to move away from foreign accelerators, The Register reported.

• Huawei is deploying a 256,000-card Atlas 950 SuperCluster, per The Register.
• Its newer architecture is designed to scale to as many as one million NPUs.
• Xu said Huawei lacks capacity to satisfy Chinese demand and has no plans for a full international push.

Washington signs voluntary AI rules

On 29 September, President Trump and the leaders of the largest US AI companies signed a set of voluntary safety standards at the White House, CBS News reported. "It's almost like a constitution, in a way," Trump said, flanked by SpaceX's Elon Musk, Meta's Mark Zuckerberg, Anthropic's Dario Amodei, Nvidia's Jensen Huang, OpenAI's Greg Brockman and Google's Sundar Pichai. The one-page accord commits the companies to "four layers of controls and audits", including internal evaluations, audits by an external firm and board reviews, according to a copy Trump posted to Truth Social. Asked whether the deal was binding, Trump said: "I think it's morally binding."

The White House meeting did not touch export controls, which remain the live dispute between Washington and Beijing. Beijing has leaned on domestic buyers instead. A 30 September essay by investor David Cheng on the Earned Intuition Substack described Chinese AI as the one sector the state has decided must work, propped up with subsidised electricity, fast-tracked IPOs and tolerance for price wars. He cited youth unemployment of 17.9 percent as of July and July retail sales up 0.6 percent year on year against an expected 1.5 percent.

Separately, the EU Council is pushing ahead with its Chat Control 2.0 scanning proposal. A leaked presidency note dated 18 September, reported by Reclaim The Net on 30 September, shows the Council wants police to be able to order scans of "parts of a service" rather than target a specific suspect. Former MEP Patrick Breyer called it "mass surveillance by another name" and said that in Germany over half of investigations target minors for consensual sexting while 75 percent of flagged chats are not actionable.

On the design-tool side, Cadence, Synopsys and Siemens EDA are all shipping agentic AI for chip design in 2026, Tom's Hardware reported on 30 September. Cadence said at Computex on 1 June that its agent reached what it calls Level 5 autonomy; Synopsys said its spec-to-RTL workflow shown on 11 March reached "L4" and now claims L5 for long-horizon agents; Siemens launched its agent on 16 March with self-verifying loops announced on 26 July. All the speed-up figures are the vendors' or their customers' own, often with "up to" caveats, Tom's Hardware noted.

Emergence AI, meanwhile, is moving its neuroformal AI into deployments with fabless chipmakers, EE Times reported on 30 September. Co-founder Satya Nitta told the outlet: "You cannot make chips any faster, but what you can do is definitely get more chips per wafer yielding with AI."

AMD, for its part, published ADLX 2.0 on 30 September, adding Python bindings and Model Context Protocol servers so AI agents can read GPU telemetry and change graphics settings, according to the company's GPUOpen blog.