DeepSeek Reportedly Trained Its New Model on Banned Nvidia Blackwell Accelerators
Chinese company DeepSeek has allegedly trained its latest AI model on Nvidia Blackwell accelerators, which remain under an export ban to China. Despite a partial easing of US sanctions policy, Blackwell-family chips still cannot be legally shipped to the PRC. If the information is confirmed, it would be a serious blow to the effectiveness of the US export control regime, and it raises the question of whether sanctions can contain China's AI development at all.
AI-processed from 3DNews AI; edited by Hamidun News
The US export control regime, built to slow the development of Chinese artificial intelligence, has run into what may be the most telling failure in its history. According to sources cited by American analysts, the Chinese company DeepSeek used Nvidia Blackwell accelerators, among the most powerful chips and strictly banned from export to China, to train its latest AI model.
To grasp the scale of the story, it helps to recall the logic behind the US sanctions. Since 2022, Washington has steadily tightened restrictions on shipments of advanced computing accelerators to China. The idea was simple: without access to the most powerful chips, Chinese companies would not be able to train competitive large language models and other AI systems.
Nvidia, which dominates the market for machine-learning accelerators, ended up at the center of this strategy. The company had to build special "cut-down" versions of its chips for the Chinese market, while its flagship products, including the entire Blackwell family, remained under a strict ban. Even though the US administration has recently softened its overall rhetoric somewhat and partially eased restrictions on less powerful products, Blackwell still falls under the strictest embargo category.
DeepSeek has demonstrated more than once that it can achieve impressive results with limited access to computing resources. The company's models, notably DeepSeek-V3 and DeepSeek-R1, hit the industry like a bombshell: they delivered performance comparable to Western counterparts at significantly lower training costs. The company then claimed it had worked on older Nvidia A100 accelerators, also formally sanctioned, but which reached China before the bans were introduced or through gray channels. Now the talk is of Blackwell, an architecture that began shipping to customers only in late 2024, meaning under restrictions already in force.
How exactly Blackwell accelerators could have ended up in DeepSeek's hands remains an open question. There are several plausible scenarios. The first is the use of intermediaries and shell companies in third countries, through which chips are rerouted to China. This channel is well documented: US authorities have repeatedly recorded cases of re-export via Southeast Asia and the Middle East. The second scenario is access to Blackwell-based cloud computing capacity through foreign data centers, which may formally not violate the letter of sanctions law but completely undermines its spirit. The third is direct illegal shipments, whose scale, according to experts, may be significantly larger than commonly assumed.
For Nvidia, the situation is highly awkward. Jensen Huang's company is already losing billions of dollars because it cannot sell its best products in one of the world's largest markets. Every such report increases pressure from Congress, which demands even tougher measures, and at the same time undercuts Nvidia's own argument that excessive restrictions hurt American companies without achieving their stated goals. The paradox is that both sides of the debate, those pushing for tighter rules and those opposing them, can use this story to their advantage.
For the global AI race, the consequences are even more serious. If DeepSeek really did train the model on Blackwell, the gap between Chinese and Western AI systems may be even smaller than assumed. A company that managed to build competitive models even on older hardware could make a qualitative leap once it gets advanced chips. This calls the entire containment strategy into question: if sanctions work not as a barrier but only as a brake, their real value is significantly lower than the political price paid for them.
This story will likely become a turning point in the debate over the future of export controls. Washington will have to acknowledge an uncomfortable truth: in the globalized semiconductor world, absolute control over supply chains is an illusion. Chips, like water, find their way to where there is demand. And as long as that demand is measured in billions of dollars, no sanctions will be able to shut off the flow completely.
- Nov 2, 2026TensorFold 1.2.0 boosts Qwen3.8-27B throughput on M3 Ultra by 44.2% at 16 streams
- Nov 1, 2026Local models lose to Opus and Codex: 0.21 on a MacBook versus 1.00 for Codex
- Nov 1, 2026Habr author fine-tunes YOLOv8l on dental X-rays and pushes mAP50 to 0.642
- Nov 1, 2026Voice bots built on LLMs turn to Jev, a fast classifier, for decisions in 200–500 ms
Test Frontier LLMs & Autonomous Agents in One Workspace
Run Claude Opus 5.5, GPT-6, Gemini 3.1, and DeepSeek side-by-side in Nomad AI (nomadnet.ai) — unified model routing, deep research, and multimodal workflows under one subscription.
The AI world, distilled — once a week
Seven stories that actually mattered, hand-picked. No noise, no reposts, no press releases.
Done! Check your inbox for a confirmation.