▼Chinese Open-Weight Models Emerge as the Non-US AI Stack

Score history

Daily conviction score, 0 to 100. Higher means the thesis is more strongly corroborated.

Now 94 · -6 since Aug 27 · ranged 94 to 100

Showing the last few days. Unlock full score history.

Why the conviction moved

• Aug 28
  Weakened

  Rather than anchoring a separate non-US stack, Moonshot AI is trying to distribute Kimi K3 through the three US hyperscalers on a revenue-share of up to 30%, per the reported talks. A leading Chinese lab choosing American cloud distribution as its route to enterprise customers cuts against a clean split into competing supply blocs, even as the Treasury blacklist warning pushes the other way.

• Aug 28
  Strengthened +3

  Tencent's WeChat team published open weights for a 2B–9B multimodal embedding family already serving production traffic, reporting 80.6 on MMEB-v3 for the 9B. Chinese labs are now supplying not just chat models but the infrastructure components — embeddings and retrieval — that a full non-US stack needs to be self-sufficient.

• Aug 27
  Strengthened +6

  Z.ai released GLM-5.3-Flash, a 320-billion-parameter open-weight multimodal model under an MIT license, after it ran anonymously on OpenRouter as "Ox Alpha" and became the platform's most used model over a weekend. A permissive license plus demonstrated blind-preference usage removes both the legal and the quality objection enterprises cite when standardizing on a Chinese stack.

• Aug 27
  Strengthened +3

  Alibaba's Qwen team previewed its next-generation Qwen4 architecture as an open model, keeping the architectural frontier — not just the weights of trailing models — on the downloadable side of the split.

Showing the last 2 days. Unlock the full record.

Source trail

• Contradicting · August 28, 2026

  Moonshot AI Seeks Up to 30% of Revenue to Put Kimi K3 on Microsoft, Amazon and Google Clouds

  Rather than anchoring a separate non-US stack, Moonshot AI is trying to distribute Kimi K3 through the three US hyperscalers on a revenue-share of up to 30%, per the reported talks. A leading Chinese lab choosing American cloud distribution as its route to enterprise customers cuts against a clean split into competing supply blocs, even as the Treasury blacklist warning pushes the other way.

  AI ML Big Data (Telegram)

• Supporting · August 28, 2026

  Tencent's WeChat Team Open-Sources a Multimodal Embedding Family Already Running in Production

  Tencent's WeChat team published open weights for a 2B–9B multimodal embedding family already serving production traffic, reporting 80.6 on MMEB-v3 for the 9B. Chinese labs are now supplying not just chat models but the infrastructure components — embeddings and retrieval — that a full non-US stack needs to be self-sufficient.

  AI ML Big Data (Telegram)