Odd Lots Odd Lots

How an American AI Company Uses Chinese Open-Weight Models

Nov 20, 2026 · 48m

Summary

Daniel Saidi, co-founder of AI marketing firm Minerva, discusses the strategic trade-offs between open-weight and closed-source models. He explains how companies use frontier models for development but fine-tune cheaper open models for operational tasks to maximize margins. The conversation covers the economics of inference, the value of proprietary data, and security risks associated with "obliterated" open models.

Topics discussed

Intro: AI business models and the DeepSeek scare Guest intro: Minerva and the value of open-source AI Defining open-weight vs. open-source models Fine-tuning for specialized tasks and cost efficiency Data sovereignty and negotiating leverage with labs Ad break: Ripple and We Feed Raw Model durability and the economics of fine-tuning Inference infrastructure and commoditization The importance of high-quality training data Cybersecurity risks and 'obliterated' models Chinese open-source capabilities and distillation Benchmarks vs. real-world task performance Unit economics: CapEx vs. OpEx for AI startups GPU rental prices and future compute capacity Corporate adoption of open models and 'ant tax' AGI, self-iteration, and the November 3rd election Outro and sponsorships

More from Odd Lots

How AI Is Forcing China to Reshape Industrial Policy
Nov 21, 2026
This Is The Problem With AI's Most Bullish Chart
Nov 19, 2026
How GTA VI Became the Most Expensive Video Game Ever Made
Nov 18, 2026
The Impossible Trilemma That Hangs Over the Nvidia Crisis
Nov 16, 2026
How Wall Street Is Bracing For the Post-Jensen Huang Nvidia Era
Nov 14, 2026
Listen ad-free on Castria