DeepSeek is working closely with Huawei Technologies to shape the design of future AI chips, The Information reported on 19 October 2026, citing two people familiar with the matter. According to the publication, Huawei engineers are incorporating DeepSeek’s requests and suggestions into design decisions “for chips that will come out next year and beyond.”
The Information described it as “a collaboration that could help China build frontier AI models without Nvidia’s technology” and headlined it as a growing challenge to Nvidia. Its article is paywalled, and the portion we could read does not include comment from DeepSeek or Huawei. Huawei’s own roadmap, published on 17 September 2026, dates its next AI chip, the Ascend 960DT, to the first quarter of 2027.
Business Pill · GPUS EXPLAINED
A short explainer on why AI runs on chips built for video games. It teaches the general idea only and says nothing about any company in this story.
The key insight: As we read it, the report describes a model developer shaping chip design from the inside, with Huawei engineers stationed at DeepSeek’s offices, rather than a buyer choosing from a supplier’s menu.
The Information wrote that the relationship “goes beyond the kinds of conversations companies have with chip suppliers about their semiconductor road maps.” It said Huawei is taking DeepSeek’s input on design decisions “such as how models calculate and move data between processors.”
“To do this work, the Huawei engineers have been stationed at DeepSeek’s offices for an extended period, one of the people said,” the publication reported. The article is by Qianer Liu and Juro Osawa, and the readable summary does not name the chips involved; the page’s own topic tags list the Ascend 950DT and the Ascend 960DT. Techmeme’s summary of the report adds that DeepSeek plans to start using the Ascend 950DT for pre-training as soon as the end of 2026; we could not read that detail in the free portion of the article.
Per-chip FP8 compute Huawei has published for each Ascend generation: 1 PFLOPS for the Ascend 950 series, 2 PFLOPS for the Ascend 960 and 4 PFLOPS for the Ascend 970 (Huawei Connect, 18 September 2025), with availability dates from Huawei Connect 2026 (17 September 2026). The 4x comparison is our arithmetic.
What Was Posted on X
DEEPSEEK AND HUAWEI ARE COLLABORATING ON NEW AI CHIP DESIGNS TO CHALLENGE NVIDIA – THE IN FORMATION
The @DeItaone account relayed The Information’s report three minutes after it was published, in a one-line headline: DeepSeek and Huawei collaborating on new AI chip designs to challenge Nvidia. It adds no detail beyond The Information’s report, which this piece quotes from the article’s free summary.
A month earlier, on 21 September 2026, The Information’s Qianer Liu posted that DeepSeek chief executive Liang Wenfeng is betting on Huawei chips and that the company expected a new batch of Huawei chips for AI training as early as the fourth quarter, summarising an article by the same two reporters who wrote today’s report.
Huawei’s Published Chip Roadmap
At Huawei Connect 2025, on 18 September 2025, Huawei said it would work on “three new series of Ascend chips: the Ascend 950 series, which includes the Ascend 950PR and the Ascend 950DT, the Ascend 960 series, and the Ascend 970 series.” It gave per-chip figures: “The Ascend 950 series will provide 1 PFLOPS in FP8 and 2 PFLOPS in FP4. The Ascend 960 will churn out 2 PFLOPS in FP8 and 4 PFLOPS in FP4. The Ascend 970 will offer 4 PFLOPS in FP8 and 8 PFLOPS in FP4.”
The Ascend 950DT, Huawei said then, would pair with memory “delivering 144 GB of memory and 4 TB/s memory access bandwidth” and “will be available in the fourth quarter of 2026.”
A year later, at Huawei Connect 2026 on 17 September 2026, Huawei said the “Ascend 960DT will be available in Q1 2027, three quarters ahead of the company’s original roadmap. And the Ascend 960PR will be ready in Q3 2027, one quarter ahead of schedule.” “We’re evolving our Ascend chip series on a one-generation-a-year cycle,” Huawei’s rotating chairman David Wang said in his keynote. “In 2028 and 2029, we will roll out the Ascend 970 and 980 chips, respectively.”
By our arithmetic from Huawei’s published figures, the Ascend 970 is planned at four times the FP8 compute of the Ascend 950. By Huawei’s published dates, the Ascend chips due in 2027, the “next year” in The Information’s report, are the Ascend 960DT and the Ascend 960PR; that pairing is our reading, not a claim in the readable report.
Huawei’s Ascend 960 was due in the fourth quarter of 2027 on its September 2025 roadmap; Huawei’s September 2026 release dates the Ascend 960DT to the first quarter of 2027, three quarters ahead. Quarters counted from Q1 2026, our arithmetic.
DeepSeek’s Trail of Chip Hints
DeepSeek has pointed at Chinese chips before. The model card for DeepSeek-V3.1 says the model was trained with the UE8M0 FP8 scale data format “to ensure compatibility with microscaling data formats.” CNBC reported on 22 August 2025 that DeepSeek said, in a comment on its official WeChat account, that the format “is tailored for the next-generation domestically built chips that will be launched soon,” and that DeepSeek “did not disclose the chips it used to train the V3.1, or what local chips the UE8M0 FP8 might be compatible with.”
On 21 September 2026, The Information’s Qianer Liu posted on X that DeepSeek chief executive Liang Wenfeng “is betting BIG on Huawei chips” and that “the company expects to receive a new batch of Huawei chips for AI training as early as Q4,” summarising an earlier article by the same reporters.
The Structural Read
As we read it, the report and Huawei’s roadmap fit together on timing: The Information says the input concerns chips that will come out next year and beyond, and Huawei’s published dates put the Ascend 960DT in the first quarter of 2027 and the Ascend 960PR in the third quarter.
As we read it, DeepSeek’s earlier statements point the same way: its V3.1 model card names the UE8M0 FP8 format, and CNBC reported that DeepSeek tied that format to next-generation domestically built chips.
As we read it, the design questions named in the report, how models calculate and move data between processors, are the same kinds of specifications Huawei publishes for each Ascend generation: data formats, memory bandwidth and interconnect bandwidth.
David Wang, Huawei rotating chairman, Huawei Connect 2026
“Weโre evolving our Ascend chip series on a one-generation-a-year cycle”
Three Implications
NEXT YEAR’S CHIPS ARE ALREADY DATED Huawei dates the Ascend 960DT to Q1 2027, three quarters ahead of its original roadmap, and the Ascend 960PR to Q3 2027.
COMPUTE DOUBLES PER GENERATION ON PAPER Huawei’s published FP8 figures go from 1 PFLOPS for the Ascend 950 to 2 for the 960 and 4 for the 970, our arithmetic: 4x.
NEITHER COMPANY HAS CONFIRMED IT The readable report carries no comment from DeepSeek or Huawei, and neither has announced the collaboration in a release we could find.
The framework puts it this way: “It is a layered industrial stack — and at any given moment, the binding constraint sits at a different layer than it did six months ago.”
As we read it, the reported DeepSeek-Huawei work joins two of those layers, the model and the chip, inside one Chinese supply chain. The framework was published in May 2026 and does not discuss this report.
What Is Not Established
The readable part of The Information’s report does not name the chips, give any financial terms, say how many Huawei engineers are involved or since when, or carry comment from DeepSeek or Huawei. Neither company has announced the collaboration in a release we could find, and we did not read the full article.
Business Engineer Framework
The Nine Layers of AI
A Business Engineer map of the AI stack from bedrock to surface, and where the binding constraint sits.
DeepSeek is working closely with Huawei to shape the design of future AI chips, with Huawei engineers stationed at DeepSeek’s offices, The Information reported on 19 October 2026, citing people familiar with the matter. Huawei’s published roadmap puts its next AI chip, the Ascend 960DT, in the first quarter of 2027, and it plans the Ascend 970 for 2028 at four times the Ascend 950’s FP8 compute, by our arithmetic.
A note on sourcing. The collaboration is reported by The Information (19 October 2026), citing two people familiar with the matter; its article is paywalled and we read only the free summary, which carries no comment from DeepSeek or Huawei. Chip figures and dates are from Huawei’s own Huawei Connect keynote (18 September 2025) and release (17 September 2026). DeepSeek’s chip-format statement is from its V3.1 model card and CNBC’s report of 22 August 2025.
The 4x comparison is our arithmetic and the pairing of next year’s chips with the Ascend 960 series is our reading. Nothing here is a forecast, and nothing here is financial or investment advice.
Gennaro Cuofano is a CRO and tech executive who has worked in AI since late 2015, starting with bringing NLP, natural-language generation, voice and chatbot products to the marketing industry. His background is in law and finance: he holds a Master's degree in Law and an International MBA with an emphasis on corporate finance (LUISS Business School and the University of San Diego, 2012), and worked as a financial analyst at a real-estate investment firm in San Diego and as an assistant controller. His work focuses on business model strategy, business engineering and, more broadly, structural analysis: how companies make money, read from their own filings. He created FourWeekMBA and leads research there and at The Business Engineer, his newsletter on AI and business strategy, with over 95,000 subscribers and more than 1,000 published analyses. His writing has also appeared on Entrepreneur (for example, How to Digitize a Family Business in Europe) and HackerNoon. Find him on LinkedIn and Substack. See how we source.
