Deepseek may be preparing another high-profile moment in the AI race. According to Reuters, citing a senior Trump administration official, the Chinese AI startup is expected to release its latest AI model next week after apparently training it on Nvidia's most powerful Blackwell chips despite the US export ban.
The report puts Deepseek back at the center of several tensions at once: access to advanced hardware, the pressure on US AI labs, benchmark credibility, and the possibility that another low-cost model could unsettle the market.
What Reuters reported about Deepseek
The core claim is specific. Reuters, citing a senior Trump administration official, reported that Deepseek has apparently trained its latest AI model using Nvidia Blackwell chips. The source article describes those chips as Nvidia's most powerful Blackwell chips.
The official said the chips are believed to be located in a data center in Inner Mongolia. The same official also said Deepseek is expected to scrub technical fingerprints of US chip usage before the model is released.
Several important details remain unresolved in the source. The official would not say how Deepseek obtained the chips. Nvidia declined to comment. Deepseek and the US Department of Commerce did not respond to Reuters.
That leaves the public picture incomplete. The reported hardware claim is significant, but the available source does not establish the supply chain, the method of acquisition, or any response from the company at the center of the allegation.
Why the Blackwell chip claim matters
The reported use of Blackwell chips matters because the source connects it directly to the US export ban. If Deepseek's latest AI model was trained on those chips, the release would raise new questions about how advanced computing power moves around formal restrictions.
The source also notes that rumors about chip smuggling had already been circulating since late last year. It does not confirm those rumors or describe specific routes, participants, or evidence. But it makes clear that the reported Blackwell connection is not landing in a vacuum.
For AI developers, chips are not a side issue. The source frames Deepseek's upcoming model as potentially powerful, and it ties that expectation to access to high-end Nvidia hardware. That is why the question of what hardware was used has become part of the story before the model itself has even arrived.
The timing also matters. The model is expected next week, according to the Reuters-based report. That means the hardware claims, the expected release, and the broader market reaction are now closely linked.
US AI labs are watching for another low-cost shock
The source says the timing of the leaks may suggest Deepseek is close to another major splash. It also says Google, OpenAI, and Anthropic have all been complaining about distillation attacks on their models by Chinese startups.
Distillation attacks are mentioned in the source as part of the pressure surrounding Deepseek and other Chinese startups. The article does not provide technical details, named examples, or specific incidents, so the safest reading is narrower: major US AI labs are publicly concerned about how their models may be used by competitors.
OpenAI is also described as having recently moved to relativize a well-known coding benchmark. The source presents that move alongside the complaints from Google, OpenAI, and Anthropic, suggesting that benchmark performance and model comparisons are becoming part of the competitive response.
Taken together, the source argues that these moves suggest Deepseek may again deliver strong results at rock-bottom prices. That phrase captures the market fear around the company: not simply that it could release a capable AI model, but that it could do so in a way that challenges assumptions about cost.
The memory of January 2025 still hangs over the market
The source explicitly points back to January 2025, when China's leading AI startup sent shockwaves through US tech stocks riding the AI bubble. That earlier reaction is part of why the next Deepseek release is attracting attention before it is public.
Markets and AI companies do not respond only to finished products. They respond to signals. In this case, the signals include the expected release next week, the reported Blackwell chip usage, the alleged data center location in Inner Mongolia, and the concern among Google, OpenAI, and Anthropic about Chinese startups.
The story is therefore not only about one model. It is about whether Deepseek can again shift expectations around AI performance and pricing. The source stops short of showing the model's capabilities, but it makes clear that several major players are acting as if the release could matter.
What remains unknown before the release
Several facts are still missing from the public account presented in the source. Those gaps are important because they limit what can be concluded before the model appears.
- How Deepseek obtained the Nvidia Blackwell chips has not been explained in the source.
- Nvidia declined to comment.
- Deepseek did not respond to Reuters.
- The US Department of Commerce did not respond to Reuters.
- The model's actual performance, pricing, and technical details are not yet described in the source.
That makes the next step straightforward: the model's release, if it arrives next week as expected, will determine how much of the current concern turns into measurable competitive pressure.
For now, Deepseek's next AI model sits at the intersection of hardware access, export controls, benchmark politics, and pricing pressure. The facts available from the source are limited, but they are enough to explain why Google, OpenAI, Anthropic, Nvidia, and US officials are all part of the same unfolding story.