Skip to content
NewA calmer, faster ZeroHour
StoryAI policyinfo

Researcher says six Chinese models show Nvidia Blackwell fingerprints

4 sources · 4 articlesFirst reported Updated ()2
What's new

November 13–16: Decoder adds the 4,500-GPU Kimi K3 estimate and remaining Hopper labs; TechCrunch cites Epoch import figures; MIT says the findings stay unconfirmed.

Merged summary · x-ai/grok-4.7 · rewritten as coverage arrivesSummaries by Vani AI

Will DePue’s Blackwell-fingerprint claims remain unconfirmed, while a US rule targets remote training on Nvidia chips.

Former OpenAI researcher Will DePue says six Chinese models released since April 2026 carry unused-weight patterns matching Nvidia Blackwell core counts. He maps DeepSeek-V4-Flash and MiniMax-M3 to 148-core B200 or B300 chips, and DeepSeek-V4-Pro, Kimi K3, Kimi K3.1-Pro, and Ling-3.1-flash to 152-core GB200 or GB300 systems. He estimates about 12,000 Blackwell GPUs across five runs, including about 4,500 for Kimi K3 over roughly 70 days, and says the method matched 24 known-hardware controls but does not locate the chips or prove an export violation. MIT Technology Review says the findings are not independently confirmed and do not show Nvidia helped. A November 16 Commerce Department rule seeks to stop Chinese firms training on Nvidia chips accessed remotely, including in Malaysia.

Key facts
  • DePue links six post-April 2026 Chinese models to Blackwell core counts.
  • Five examined runs estimated at about 12,000 Blackwell GPUs combined.
  • Kimi K3 alone estimated at about 4,500 Blackwell GPUs over 70 days.
  • Qwen3.8 and Zhipu, Tencent, Xiaomi, and StepFun models still match Hopper.
  • DePue says the method matched all 24 known-hardware control models.
  • November 16 Commerce rule targets remote Nvidia-chip training, including Malaysia.
Explain this AI news with Vani AIWhat was announced, what is new and why it matters, in a new chat

Timeline

dated events in the reporting · oldest first
  1. Nov 13, 2026DePue publishes Blackwell fingerprint claims and verification code. · The Decoder
  2. Nov 16, 2026Commerce Department rule targets remote training on Nvidia chips. · MIT Technology Review

What each outlet adds

beyond the shared facts
  • The DecoderModel-to-core mapping, 12,000-GPU and Kimi K3 estimates, Hopper holdouts.
  • TechCrunchKratsios claim, Epoch import estimates, and Hong Kong H200 stockpile.
  • MIT Technology ReviewFindings unconfirmed; Monday remote-training rule; petition after Huang resigned.

Coverage

oldest first · one row per article · UTC
  1. · 6d ago The DecoderFirst report

    Researcher says six Chinese models show Nvidia Blackwell fingerprints

    Will DePue says unused weights preserve Hopper or Blackwell core-count patterns and published verification code.

  2. · 6d ago TechCrunch · AI

    Research claims banned Nvidia chips trained China’s leading open models

    A former OpenAI researcher says Kimi K3 and DeepSeek V4 weights show Blackwell fingerprints; Jensen Huang left Nvidia this week.

  3. · 5d ago Interconnects

    Author says Huang exit endangers US open-model support

    The essay argues Nvidia open-model funding will likely dry up as the company prioritizes stabilizing deliveries for 24 months.

  4. · 3d ago MIT Technology Review · AI

    Commerce rule targets remote Nvidia chip use as Huang petition grows

    Unconfirmed research alleges Chinese AI firms had wider access to Nvidia Blackwell chips than previously estimated.