Forensic Analysis Suggests Chinese AI Labs Are Training Models on Restricted Nvidia Blackwell Chips
A new forensic report claims that top Chinese AI labs have used restricted Nvidia Blackwell hardware for model training, highlighting potential loopholes in current US export controls.
🔌Chips & HardwarePhoto: theregister.com
Key points
- Forensic analysis of model weights reveals signatures matching Nvidia Blackwell GPUs.
- Affected developers include DeepSeek, Moonshot AI, MiniMax, and Ant Group.
- Experts note that hardware usage does not definitively prove the physical location of the chips.
Prominent Chinese artificial intelligence labs, including DeepSeek, Moonshot AI, MiniMax, and Ant Group, have reportedly utilized restricted Nvidia Blackwell graphics processing units to train their models. According to a forensic report published by Will DePue, a former OpenAI engineer and founder of research consultancy Unemployment Labs, published model weights contain hidden signatures that reveal the underlying hardware used during initialization.
DePue's analysis focused on how initial weights are generated using pseudorandom number generation algorithms like Philox alongside Nvidia's cuRAND library. By scanning millions of combinations of chip layouts and seed numbers, the researcher matched open-weight models from Chinese developers to Nvidia Blackwell architectures, including flagship liquid-cooled rack-scale systems like the GB200 and GB300. The report notes that while the findings strongly indicate the use of Blackwell hardware, they do not definitively prove the physical location of the chips, meaning some labs may have accessed the technology remotely via US cloud providers.
The disclosure arrives amid intensified scrutiny in Washington over the efficacy of export restrictions designed to keep advanced semiconductor technology out of China. Lawmakers are currently weighing multiple legislative measures, such as the AI Overwatch Act and the Cloud Security Act, to tighten oversight and close existing gaps in infrastructure access.
What to watch next: Congressional movement on pending export control legislation and potential responses from federal regulators regarding cloud-based hardware access.
Source theregister.com — theregister.com/ai-and-ml/2026/11/13/chinese-ai-labs-are-training-m… ↗
Read next
🚀Startups & FundingHarvard Study Reveals US Hyperscale Data Centers Consume Power Rivaling Entire States
A new Harvard study reveals that America's 403 hyperscale data centers consumed between 68 and 99 terawatt-hours of electricity during the 2024-25 study period, rivaling …
🚀Startups & FundingCalifornia Files Lawsuit Against OpenAI Over Autonomous AI Incidents
California Attorney General Rob Bonta has filed a lawsuit against OpenAI, alleging the company failed to control its autonomous AI agents and misled the public about safe…
🧠AI AnalysisPedestal AI Secures Seed Funding to Automate Logistics With Intelligent Agents
Pedestal AI has launched publicly with $4.7 million in seed backing to deploy agentic artificial intelligence aimed at streamlining logistics and supply chain operations.
🏢Big TechUnited Launch Alliance Investigates Vulcan Rocket Failure
United Launch Alliance is investigating a mission failure of its Vulcan rocket, which suffered from low booster performance shortly after liftoff. The incident raises que…
🌐Work & SocietyUS Southwest Hydropower Output Hits Record Lows Amid Persistent Drought
Hydropower generation in the US Southwest dropped 28% below the 10-year average this summer as record-low water levels in the Colorado River Basin threaten regional energ…
🏢Big TechYouTube Updates View Metrics to Include TV Co-Viewing Estimates
YouTube is incorporating estimated co-viewing data into its public view counts to better reflect group consumption on television sets.