Big Technology Podcast · Alex Kantrowitz

Who Wins if AI Models Commoditize? — With Mistral CEO Arthur Mensch

·57 min·3 clips

“Treat intelligence as electricity” and avoid having your access throttled, Mensch says about open source AI.

1. Big Technology Podcast asks what the AI business looks like when leading models start to look the same. 2. Host Alex Kantrowitz speaks with Arthur Mensch, the CEO and co-founder of Mistral, whose company is based in France and was valued at $14 billion after starting in April 2023. 3. The episode asks whether commoditization pushes value away from frontier models and toward enterprise software, open source, and implementation services. 4. Mensch says AI will be commoditized because about 10 labs know how to build frontier systems, the training recipes are short, and similar data sources circulate quickly. 5. He argues that the hard problem is not just building models, but finding a business model that can stay profitable when model assets depreciate quickly. 6. The conversation turns to OpenAI’s reported $1.4 trillion infrastructure ambition and whether model builders should keep spending at that scale. 7. Mensch says enterprises need value fast enough to justify the industry’s collective spending. 8. He gives the example of supply-chain workflows where a team of 20 people might be replaced by 2 people through better AI customization. 9. He says companies often focus on “the solution” instead of starting from the problem they actually need to solve. 10. The discussion shifts to AGI, and Mensch says the concept is too simple for enterprises and does not match the complexity of real-world systems. 11. He describes AI products as a mix of static definitions set by humans and dynamic execution where models call tools and choose paths. 12. He says orchestration, guardrails, and agentic behavior all remain important because a fully dynamic system without human guidance is unrealistic. 13. Mensch says AI business value splits into consumer chatbots, enterprise re-platforming, and specialized products built around proprietary data. 14. He argues that enterprise software will be re-platformed around unified data, context engines, and generative interfaces that produce the right widgets on demand. 15. He also says AI can unlock technological progress in domains such as manufacturing, physics, ASML-style machine operation, nuclear fusion, and semiconductor engraving. 16. On open source, Mensch says closed-source and open-source model gaps are shrinking because 10^26 flops is enough for many labs to reach comparable training scale. 17. He says Mistral’s latest open release, DevStyle 2, is performing around the level of Anthropic models from two or three months earlier. 18. The interview keeps a measured, technical tone, with long answers from Mensch and pointed follow-up questions about business models, control, and vendor lock-in. 19. Listeners interested in AI strategy, enterprise software, and open source competition will get the most from it. 20. Listeners looking for a fast product demo or a light technical overview may skip it.

Listen to the show on

Apple Podcasts Spotify

Play clips from this episode More from Big Technology Podcast