Medium impact

Top Chinese AI labs built their newest models on Nvidia's Blackwell chips

NVIDIA•USA•

What happened

A forensic analysis of released AI model weights shows that four major Chinese AI labs—DeepSeek, Moonshot, MiniMax, and Ant Group—trained their newest models on NVIDIA’s Blackwell GPUs, which are export‑controlled and barred from China. The study matched over six trillion GPU‑seed combinations to model fingerprints, finding strong evidence that models such as DeepSeek‑V4‑Pro, Kimi K3, and MiniMax‑M3 were initialized on Blackwell hardware. The analysis was independently verified by researchers at OpenAI and Anthropic. The findings suggest that these labs may have spent hundreds of millions of dollars on prohibited GPUs.

Why it matters

If true, it indicates that Chinese AI firms are accessing advanced hardware that should be restricted, potentially giving them a competitive edge. This could raise concerns about the effectiveness of export controls and the global balance of AI development.

Reality check

An automated, sceptical read of the source

Genuine breakthrough. The claim that export‑controlled GPUs were used by major Chinese labs, supported by extensive forensic evidence, represents a significant and potentially game‑changing revelation.

Who it affectsResearchersPolicymakers
What to watchFuture reports may reveal additional models or confirm the scale of GPU usage by these labs.
Do you agree with this rating?

Is this a big deal?

Get stories like this every morning

The day's AI news in plain English, in your inbox at 8:30 IST. Free.

More from NVIDIA

How AI Is Helping Care Teams Get Patients the Right Care, SoonerNVIDIA · 13 Nov 2026A New Dawn in the Cloud: ‘Warhammer 40,000: Dawn of War IV’ Is Coming to GeForce NOWNVIDIA · 12 Nov 2026Nvidia Faced Pressure to Boost Its AI Backstop Before Huang ExitNVIDIA · 12 Nov 2026Washing dishes at Denny's to rockstar of AI, the rise and fall of Jensen HuangNVIDIA · 12 Nov 2026How AI Helps Recyclers Turn Mixed Waste Into Reusable MaterialNVIDIA · 11 Nov 2026

This page was collected from twitter.com and summarised automatically by AIPulsee's pipeline. See how it runs.