Chip export controls tighten around a market Huawei says it now leads

The claim appeared in a Q&A-style release and was reported by The Register on 30 September. It cannot be verified from outside Huawei, and Xu said as much himself. "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," he said. That caveat matters. The number is the whole story. If it holds, the export control regime has converted Nvidia's most important growth market outside the US into a domestic supply chain problem in roughly eighteen months.

The policy timeline is not in dispute. In April 2025, the US Commerce Department imposed new licensing requirements on Nvidia's and AMD's China-spec accelerators. That summer it reversed course, and later it reached an arrangement to take a percentage of certain China-bound chip sales, The Register reported. By late last year, the Trump administration announced that Nvidia would be allowed to sell its more potent H200-series parts to approved Chinese customers for the first time. Nvidia's Q2 earnings call revealed that H200 shipments to the region amounted to less than 1 percent of datacenter revenues.

Self-sufficiency as a policy, not a slogan

Xu is blunt about why. "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. "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."

Huawei is backing that with hardware at a scale that is hard to dismiss. According to The Register, the company is currently deploying a 256,000-card Atlas 950 SuperCluster. 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. On paper, Huawei's accelerators still trail Nvidia and AMD. In practice, the argument goes, scale and guaranteed supply make up the deficit.

Software is the other half of the problem, and here the picture has moved quickly. DeepSeek, working with Huawei, released open-source programming tools for Ascend chips. The centerpiece is TileLang, a language intended to be easier to work with than Nvidia's CUDA platform, The Decoder reported on 30 September, citing Reuters.

"The first barrier encountered by Ascend is actually not hardware itself," said Liao Heng, chief scientist of Huawei's HiSilicon. "It's primarily the ecosystem barrier, because two years ago there was still a tremendous software barrier for Ascend."

Liao is pointing straight at the CUDA moat, the four million or so developers Nvidia has accumulated over more than a decade. That moat is not unique to China, but it is the one Beijing has spent the most money trying to cross. According to The Decoder, DeepSeek says TileLang offers a simpler programming model than CUDA. Researchers at Peking University originally developed it, and it has already been tested on older Nvidia chips. The New York Times reports it is now DeepSeek's main tool for its AGI work.

Independent analysis suggests the moat is at least weakening. SemiAnalysis tested OpenAI's Jalapeño inference chip and called the CUDA moat "potentially dead," because OpenAI gets new models running on its own hardware so quickly. Jalapeño beat Nvidia's Blackwell on performance per watt in most scenarios tested. The analysts attached caveats. They tested relatively easy-to-optimize scenarios of about 8,000 input tokens and 1,000 output tokens, and did not run AgentX, the benchmark for multi-step agent tasks. Their conclusion is not that Nvidia's silicon has lost its edge, but that its lasting advantage sits in the software that links many chips into one system.

The enforcement problem underneath

All of this would be a straightforward story about industrial policy working as intended, if restricted US silicon were not still turning up in Chinese designs. On 24 September, Tom's Hardware reported that Dylan Patel, head of SemiAnalysis, had called for a treason investigation into AMD. Patel found that China's Puzhi Electronic Technology was crowdfunding a system built around AMD's Zynq UltraScale+ XCZU47DR RFSoC, a US export-controlled dual-use component.

Patel's complaint was as much about price as about provenance. "U.S. list price is $36K for this chip, with $4K – 5K to U.S. companies at volume, but it is getting quoted $1K in China to crowdfunding campaigns," he wrote on X. "AMD needs to be investigated for treason."

AMD denied selling or shipping the chips directly into the restricted region. "AMD is committed to full compliance with all U.S. and global export regulations and has a clear policy requiring our distributors, resellers, and customers to do the same," a spokesperson told Tom's Hardware. "Any shipment into restricted regions is a direct violation of our policy. This recent instance is unrelated to any direct AMD sales or shipments. We investigate all suspected diversion cases and if we find any violation of law, we will report it to relevant authorities and cease all business with the company."

The technical detail explains why this class of part is sensitive. Tom's Hardware reported that the Zynq UltraScale+ XCZU47DR packs four general-purpose Arm cores, two Cortex-R5F real-time cores, an FPGA with 930,000 gates, and DSP engines with 4,272 DSP slices, plus eight 14-bit ADCs capable of up to 5 GSPS and eight 14-bit DACs running at up to 9.85 GSPS. Devices in the family go into software-defined radios, radar and phased-array systems, satellite communications and electronic test equipment. Puzhi positions its systems for "professional engineers" and "advanced researchers in universities."

No public evidence shows that this particular case involved AMD selling directly into China, and the company did not explicitly confirm to Tom's Hardware that it had opened an investigation into this instance. What the episode illustrates is narrower and more uncomfortable. Export controls govern the point of sale, and dual-use silicon can move through third-party channels afterwards. Officials have spent two administrations trying to close that gap, with mixed results.

Design tools become the next contested layer

While the enforcement argument grinds on, AI partnerships outside the export control debate are reorganising the design side of the industry. On 30 September, OpenAI and Synopsys announced a multi-year strategic partnership to build a specialised chip design model called GPT-Synopsys, according to The Decoder. OpenAI is licensing Synopsys' EDA tools for the project. The stated goal is a model that can reason about chip design and verification and directly operate Synopsys' tools, with engineers delegating objectives and approving output. Both companies say customer data will not be used for training and will be stored encrypted, and that early tests with semiconductor customers are already underway.

The broader tooling market is moving in the same direction. Tom's Hardware's AI Chip Design Week, running from 28 September to 2 October, notes that Cadence, Synopsys and Siemens all now offer agentic AI for chip design, largely built on Nvidia's stack, with varying claims of autonomy. Synopsys unveiled seven AgentEngineer agents on its new Autopilot platform, with general availability planned for the end of 2026.

None of this resolves the export control question, and it may make it harder. If AI-assisted design tools compress the time and expertise needed to build complex silicon, the practical barrier to entry falls for everyone, including the Chinese firms Washington has been trying to slow down. Huawei's own position remains limited by capacity rather than policy. "We don't have enough capacity to satisfy the demand in China," Xu said. "We don't have plans to expand the international market in a fully-fledged way."

That may be the most accurate description of where the control regime has landed after eighteen months of reversals. It has not stopped China's domestic chip industry, and it has not preserved Nvidia's position there either. It has produced a market where supply certainty, not performance, is the selling point, and where the enforcement cases that still surface arrive through crowdfunding pages rather than corporate filings.