ChinaTalk Overfit is now ModelTalk! GPU Smuggling, OpenAI Cooked? + Open Models, AI Writing
47 snips
Mar 23, 2026 
Jasmine Sun, a journalist who covers AI, labor, and tech economics, and Nathan Lambert, an AI analyst and co-founder focused on GPUs and infrastructure, trade sharp takes. They dive into GPU smuggling and export-control motives. They compare GPT-5.4’s style, debate open versus closed models, and unpack AI writing, local models, and researcher incentives.
AI Snips
Chapters
Transcript
Episode notes
Supermicro's Billion Dollar GPU Smuggling Example
- Supermicro was a major real-company example in the GPU smuggling case that stunned the hosts.
- Jordan described $2–3 billion of chips routed through Singapore with fake dummy racks to hide shipments into China.
GPT 5.4 Is Smarter But Less Engaging
- GPT‑5.4 advanced intelligence but feels 'cold' and less conversational than Claude, making product adoption tricky.
- Nathan and Jordan noted ChatGPT's scale (millions of free users) is a monetization and GPU-cost burden that shapes product choices.
Don't Count On Consumer Scale Alone To Monetize
- Avoid relying solely on consumer scale for monetization; consider enterprise paths or commerce/ads as the only scalable returns.
- Jasmine and Nathan warned OpenAI that ads or commerce could drive users to alternatives if mismanaged.