#40
2026-07-01
Duo
The Real Danger Doesn't Carry an 🔓 Open Source Label!
Key Takeaways
- US regulation becomes a weapon. 100 percent digital tariffs as an answer to the European acts, plus a first regulatory intervention on frontier models, put token-based, heavily subsidised business models under refinancing pressure.
- Small models beat large ones where it counts. Hardened industry models with source, error code and diagnosis give more reliable answers for proprietary industrial data and edge deployment than a frontier model does.
- Value comes from architecture, not model size. Orchestration, guardrails and hardening around the model decide whether a system is deployable at all in a B2B setting.
- Legacy IT cannot be replaced, only hollowed out. Hollowing it out module by module from the inside preserves its book value and avoids an expensive wholesale replacement.
- Europe has a window, not a timing problem. New procurement logic in Paris or Estonia opens purchasing up to small providers as well.
- Talent hotspots such as Station F or King's Cross pull in additional people and capital.
- The greater danger is concentration of power, not open source. In Krtil's assessment, a handful of providers holding billion-euro, decades-long state contracts is the real risk — not a superintelligence emerging from open models.
Summary
In this episode of DEEPTECH DEEPTALK, Oliver Rößling and Alois Krtil assess what it means when the American state suddenly puts its own frontier models on a leash. For years the USA sold itself as the place where AI runs without brakes. Now the first regulatory interventions stand alongside the announcement of 100 percent digital tariffs as an answer to the European acts. That hits token-based, heavily subsidised business models right at their refinancing. Europe could be the beneficiary. Models such as Mistral are gaining ground, also as protection against a takeover by American or Chinese buyers. At the heart of the episode is the argument that it is not model size that decides the outcome.