AI
“Chinese AI startup DeepSeek is reportedly working with Huawei to develop AI chips that could challenge Nvidia.”
Plain restatementDeepSeek is reported, on the basis of unnamed sources, to be collaborating with Huawei on the design of future Ascend AI chips, with Huawei engineers working at DeepSeek's offices, with the aim of training advanced models on Chinese-made hardware instead of Nvidia hardware, including possible pre-training on the Ascend 950DT by the end of 2026.
This post accurately relays a real news report, and it correctly says "reportedly" and names its source. On October 19, 2026, The Information reported that DeepSeek is helping shape the design of Huawei's next generation of AI chips, with Huawei engineers working inside DeepSeek's offices, and that DeepSeek may begin large-scale training on Huawei's Ascend 950DT chip as early as the end of 2026. That report is based on two unnamed sources. Neither DeepSeek nor Huawei has confirmed or denied it, and no official announcement or document exists, so the collaboration itself remains unconfirmed. What is confirmed on the record is a narrower software partnership: DeepSeek said in late September 2026 that it had built programming tools for Huawei's Ascend chips. Two pieces of context the post leaves out are worth knowing: a reported earlier attempt to train a DeepSeek model on older Huawei chips did not produce a completed training run, and reporting from September 2026 said DeepSeek's large Huawei chip order was meant for running models rather than training them. The phrase "develop AI chips" also oversells DeepSeek's role slightly, since the reporting describes DeepSeek influencing Huawei's design choices rather than building chips.
[drifted from the evidence:] Chinese AI startup DeepSeek is [drifted from the evidence:] reportedly working with Huawei [drifted from the evidence:] to develop AI chips [drifted from the evidence:] that could challenge Nvidia.
DeepSeek is [added by the neutral restatement:] reported, on the basis of unnamed sources, to be collaborating with Huawei [added by the neutral restatement:] on the design of future Ascend AI chips, [added by the neutral restatement:] with Huawei engineers working at DeepSeek's offices, with the aim of training advanced models on Chinese-made hardware instead of Nvidia [added by the neutral restatement:] hardware, including possible pre-training on the Ascend 950DT by the end of 2026.
Red-tinted words in the claim drifted from the evidence. Green-tinted words are what a neutral restatement needs.
The trace / claim to source
- A report of this kind exists and is correctly attributed. The Information published it on October 19, 2026, and the post credits The Information.
- The post uses the word "reportedly," which matches the evidentiary status of the claim.
- The specific details in the caption track the reporting: Huawei engineers working at DeepSeek's offices, DeepSeek feedback feeding into chip design, the Ascend 950DT, and a possible start of training use by the end of 2026.
- The aim described, training advanced models on Chinese-made hardware rather than Nvidia hardware, matches how the original outlet framed the story.
- A real, publicly acknowledged DeepSeek and Huawei collaboration exists at the software level. DeepSeek itself announced programming tools optimised for Ascend chips on September 30, 2026.
- The caption says DeepSeek is working with Huawei "to develop AI chips," which reads as joint chipmaking. The reporting describes DeepSeek influencing Huawei's design decisions and supplying model requirements while Huawei designs and produces the silicon. The post's own body text describes this more accurately, so the gap sits mainly in the headline phrasing.
- The post does not say that the report rests on two unnamed sources and that neither DeepSeek nor Huawei has confirmed or denied it. A reader cannot tell from the post how thin or thick the sourcing is.
- The post does not mention that a reported attempt to train an earlier DeepSeek model on older Huawei Ascend chips, also with on-site Huawei engineers, did not produce a completed training run and that training reverted to Nvidia. That history bears directly on whether the year-end plan will work.
- The post does not note that reporting six weeks earlier said DeepSeek's large Ascend 950DT order was intended for inference while training stayed on Nvidia. The new report describes a change of plan, and presenting only the new plan removes that tension.
- Whether the chip co-design collaboration is actually happening as described. No party has confirmed it on the record and no primary document exists.
- Whether DeepSeek will in fact begin pre-training on Ascend 950DT by the end of 2026. This is described as a plan whose timing depends on testing progress.
- Whether Tencent Tech's report is an independent reporting chain or a relay of the same original sourcing.
- Whether any chip shaped by this collaboration will perform competitively against Nvidia hardware. That is a forward-looking assertion that no current evidence can settle.
- Whether DeepSeek or Huawei were asked for comment and declined, or simply did not respond. Neither a denial nor a confirmation was found.
The Information published an exclusive on October 19, 2026 stating that DeepSeek is helping shape the design of Huawei's next generation of AI chips. Its own account summarised the piece this way: "DeepSeek is helping Huawei design its next generation of AI chips, with Huawei engineers working inside DeepSeek's offices," adding that the collaboration could help China train advanced models without relying on Nvidia. The report cites two people familiar with the matter, describes Huawei engineers stationed at DeepSeek's offices for an extended period, and says their work feeds DeepSeek's requests into design decisions for chips arriving next year and beyond. The outlet wrote that the relationship "goes beyond the kinds of conversations companies have with chip suppliers about their semiconductor road maps," covering design decisions such as how models calculate and move data. Secondary summaries of the same report add the specific chip and timing details that appear in the post's caption. DeepSeek plans to begin pre-training on the Ascend 950DT as soon as the end of 2026, Huawei engineers are stationed inside DeepSeek's offices folding its feedback into how Ascend chips handle calculations and data transfer, and Huawei will prioritise DeepSeek for most of its next-generation Ascend 960DT chips. A Chinese-language relay reports the same points with sourcing of its own. Tencent Tech, via PANews on October 20, reported that unnamed sources said Huawei engineers have been stationed long-term at DeepSeek's offices and that DeepSeek plans large-scale pre-training on the Ascend 950DT as early as the end of this year, with a full pre-training run possibly requiring more than 10,000 chips running for weeks. Separate on-record evidence supports a real and public DeepSeek and Huawei software relationship. Reuters reported on September 30, 2026 that DeepSeek said it had partnered with Huawei to develop programming tools optimised for Huawei's Ascend chips, as Chinese tech firms seek alternatives to the Nvidia ecosystem. Earlier reporting points in a different direction on training. Bloomberg reported in September 2026 that DeepSeek plans to deploy at least 160,000 Ascend 950DT chips at a gigawatt-class data center for inference only, with training remaining on Nvidia, and noted that neither DeepSeek nor Huawei commented on record. There is also a documented history of difficulty. Reporting cited by Zero Hedge states that DeepSeek spent months attempting to train its R2 model on older Ascend chips with on-site Huawei engineers and, according to the Financial Times, the effort did not produce a successful training run, after which DeepSeek reverted to Nvidia for training.
Complete reasoning
The reply is formatted for pasting into the thread where the claim is circulating.
Ask this case
Answers come only from the case file above; nothing is added.
Is it true that DeepSeek is working with Huawei on AI chips?
A credible report says so, but it is unconfirmed. The Information reported on October 19, 2026 that DeepSeek is helping shape Huawei's next-generation Ascend chip designs, based on two unnamed sources, and neither company has confirmed or denied it.
Does this mean DeepSeek is building its own chips?
No. The reporting describes DeepSeek influencing Huawei's design decisions and feeding in requests about how chips should handle calculations and data, while Huawei still designs and makes the chips. The claim that DeepSeek is working 'to develop AI chips' overstates DeepSeek's role.
Is there any confirmed, on-the-record relationship between DeepSeek and Huawei?
Yes. DeepSeek said on September 30, 2026 that it had partnered with Huawei to build programming tools optimized for Huawei's Ascend chips. That software partnership is confirmed, separate from the unconfirmed chip-design collaboration.
Will DeepSeek actually train its models on Huawei chips instead of Nvidia?
This is uncertain. The report says DeepSeek may begin large-scale pre-training on the Ascend 950DT as early as the end of 2026, but this is described as a plan dependent on testing progress. Earlier reporting from September 2026 said DeepSeek's large Ascend chip order was meant for running models, not training them, and training would stay on Nvidia.
Has DeepSeek tried this kind of Huawei chip training before?
Yes, and it reportedly did not work out. According to reporting cited in the case file, DeepSeek spent months trying to train its R2 model on older Ascend chips with on-site Huawei engineers, but the effort did not produce a successful training run, and DeepSeek reverted to Nvidia for training.