Table of Contents
For years, Chinese AI labs have depended on Nvidia hardware to build their most advanced models. A new report suggests that dependence may be loosening. DeepSeek, the Hangzhou-based AI lab, is reportedly doing more than buying and programming processors. It is helping decide what goes into them, and the DeepSeek partnership with Huawei could become one of the most important stories in global technology.
What the Report Says About DeepSeek and Huawei
The Information reported on Monday, 19 October, citing two people familiar with the matter, that DeepSeek is working closely with Huawei Technologies to shape the design of future AI chips. The outlet said the collaboration could help China build frontier AI models without Nvidia technology.
It also said the relationship goes further than the usual conversations between a chip supplier and its customers. DeepSeek appears to have moved beyond writing software for Huawei’s processors and is now influencing the hardware itself.
The timing matters. US export controls limit China’s access to advanced chips and chipmaking equipment, which gives Chinese companies strong reasons to build a domestic alternative. The DeepSeek relationship with Huawei is one of the clearest signs of that effort.
From Software Partner to Chip Co-Designer
Open-Source Tools for Ascend
The two companies have been drawing closer for months. On 30 September, DeepSeek released open-source software for Huawei’s Ascend accelerators, including TileLang, a programming language meant as an answer to Nvidia’s CUDA platform, The Edge Malaysia reported.
DeepSeek said Huawei gave “unreserved and vigorous support” during development and that the two would work together on more projects. The companies also built a “supernode” system using 128 Ascend 950 chips, according to The Daily Star.
Models Adapted to Ascend
In April, DeepSeek previewed its V4 model, adapted to run on Ascend chips. Huawei said its processors were used for part of the model’s training, DD News reported.
In September, Bloomberg reported that DeepSeek plans to deploy at least 160,000 Ascend 950DT chips at a data centre it is building in Inner Mongolia, The Edge Malaysia reported. Those chips would run models rather than train them, and DeepSeek has so far relied on Nvidia accelerators for training.
That detail is worth remembering. Running a finished model, known as inference, is a very different job from training one, and training remains the harder test for DeepSeek.
Huawei’s Accelerated Roadmap
Huawei has also moved its chip schedule forward. At Huawei Connect in September, the company said the Ascend 960DT will be available in Q1 2027, three quarters ahead of its original plan, and the 960PR in the third quarter.
Rotating chairman David Wang said Huawei is “evolving our Ascend chip series on a one-generation-a-year cycle,” CRN Asia reported. The Ascend 970 is planned for 2028 and the 980 for 2029.
A yearly cycle gives a customer such as DeepSeek regular chances to feed lessons back into the next design. That is exactly the kind of loop the co-design report describes. Faster generations also mean the software tools built around them must keep pace, which makes DeepSeek’s open-source work on Ascend even more relevant.
Supply Limits Remain
Production capacity is still a serious bottleneck for the company, even as demand keeps rising. According to Bloomberg’s sources, shortages of high-end memory will limit output of Huawei’s 950DT to the low hundreds of thousands this year, and filling DeepSeek’s order could take more than a year, The Edge Malaysia reported.
Huawei has also said it cannot fully meet domestic demand for AI computing equipment, Sinolytics noted.
So even if the design work goes perfectly, the number of chips that can actually be built will shape how quickly DeepSeek can scale on Huawei hardware.
The Test Still Ahead
The partnership’s central test is still unmet. Huawei and DeepSeek have yet to show that Chinese hardware can reliably train advanced models at a speed and cost that compete with Nvidia’s.
Until they do, Nvidia keeps a meaningful lead in the part of AI development that matters most. Speed and cost are the key measures here, because a chip that trains reliably but slowly or expensively would still leave Chinese labs at a disadvantage. A strong showing on inference would be useful, but it would not prove that China can build frontier models on its own stack.
What This Means for Everyday Users
Most people will never buy an Ascend chip, yet the outcome could still reach them. If the DeepSeek and Huawei effort matures, China’s AI companies may gain more freedom to release models and services without relying on foreign suppliers. That could change pricing, availability, and competition in AI tools worldwide.
For businesses, the practical lesson is to watch the hardware layer as closely as the model layer. Software support, supply, and cost per unit of computing power all influence which AI tools become widely available. Start tracking announcements on Ascend availability and open-source tooling such as TileLang, because they are early signals of how far the shift has come.
Frequently Asked Questions
1. What did The Information report about DeepSeek and Huawei?
It reported, citing two people familiar with the matter, that DeepSeek is working closely with Huawei to shape the design of future AI chips.
2. What is TileLang?
A programming language, released by DeepSeek on 30 September as open-source software for Huawei’s Ascend accelerators, meant as an answer to Nvidia’s CUDA platform.
3. When will the Ascend 960DT arrive?
Huawei said it will be available in Q1 2027, three quarters ahead of the original plan.
4. What is holding the partnership back?
Production limits, including shortages of high-end memory, and the lack of proof that Chinese hardware can train advanced models competitively with Nvidia.
Conclusion
The reported collaboration marks a shift from software support to hardware influence. Huawei’s faster roadmap and DeepSeek’s open-source work show real momentum, yet supply limits and the unproven training question remain. Whether DeepSeek can match Nvidia on Huawei chips will decide how far China’s AI independence can go.