What happens if China pulls the plug on open-source AI? | Ep 21

July 9, 2026

AI Summary

5 min read

What happens if China pulls the plug on open-source AI?

The best open-source AI models are now coming from China. GLM 5.2, DeepSeek, Qwen, Kimi, and Minimax have all been catching up to American frontier models—GLM 5.2 is roughly GPT-4o level on agentic tasks, according to Arena's leaderboard data. But Reuters recently reported that China may stop releasing its AI models to the rest of the world. If that happens, a lot of startups that have built their businesses on Chinese open-weight models would suddenly lose their foundation.

The open-source dependency

Anastasios from Arena explained that the immediate effect would be painful: "The best open source models are coming from China. And so it's going to hurt the AI strategies of tons and tons of companies that are building with an open source first mindset." If you remove Chinese models from the leaderboard, the next best open-source options drop significantly—you're looking at Nvidia's Nemotron family, Mistral, Google's Gemma, and Reflection AI. None of them are currently at the same performance level.

Continue reading the full summary in the app — free to try.

Read Full Summary →

Free • No credit card required

Listen to Audio Summary Open in App

Never miss an episode of This Week in AI

Get every new episode summarized in your inbox — free, ~5 minutes to read.

No spam. Unsubscribe anytime.

What you'll learn

1 Timestamped Outline
2 (00:00) **China may end open-weight AI model releases** - Two startup founders react to Reuters report that China could stop releasing its frontier open-source models to the world
3 (04:04) **Cost math: why open-source is irreplaceable for multi-model architectures** - Munjal breaks down the unit economics that make closed-source models unworkable for safety-critical voice AI
4 (08:40) **Why small language models are a fallacy** - Munjal explains why only large models work for healthcare safety, and how Hippocratic uses models from 7B to 1T+ parameters
5 (10:46) **How close are Chinese models to the US frontier?** - Anastasios presents Arena data showing the gap has nearly closed, especially in agentic performance
6 (14:53) **Winners and losers of a Chinese model ban** - The strategic game theory of cutting off open-source access
7 (16:45) **The missing business model for open-weight AI** - Why open-source software economics don't translate to open-weight models

+ Full timestamped outline available in the app

Show Notes

This Week In Startups is made possible by:

PAYPAL

Today’s show:

China is weighing a ban on foreign access to its top AI models… the same low-cost open-weight systems on which half of the Valley’s startups are building. So we asked our two all-star panelists — Arena CEO/Co-Founder Anastasios Angelopoulos and Hippocratic AI CEO/Co-Founder Munjal Shah — what happens to US companies when their favorite cheap, powerful models vaporize overnight.

PLUS why Hippocratic runs 31 models in parallel to ensure clinical-grade safety, the latency-vs-intelligence tradeoff no one is optimizing for, and why a company’s benchmarks might become their most valuable IP.

Guests:

Anastasios Angelopoulos on X: https://x.com/ml_angelopoulos

Arena AI: https://arena.ai/

Agent Arena: https://arena.ai/leaderboard/agent

Munjal Shah on X: https://x.com/munjalshah

Hippocratic AI: https://hippocraticai.com/

Relevant Links:

Reuters: “Beijing is looking at curbing overseas access…”: https://www.reuters.com/world/beijing-is-looking-curbing-overseas-access-chinas-top-ai-models-sources-say-2026-07-07/

Z.ai: https://z.ai/

Alibaba’s Qwen: https://qwen.ai/

Google’s Gemma: https://ai.google.dev/gemma

Reflection AI: https://reflection.ai/