Podcast
Physical AI That Moves the World — Qasar Younis & Peter Ludwig, Applied Intuition
Latent Space: The AI Engineer Podcast
- Why Physical AI Is Bottlenecked by Hardware, Not Models The surprising constraint in autonomy is no longer model intelligence — it’s deployment. Peter Ludwig says physical AI is “not really constrained right now by the intelligence of the models”; the hard part is getting them onto real machines under brutal limits on latency, power, cost, safety, and reliability. That is why Applied Intuition built far beyond autonomy models into a full stack of simulation, operating systems, and embedded AI. Offboard models can be huge and slow, but onboard systems need answers in milliseconds, and “every fraction of a millisecond counts.” This is also why a vehicle needs a real AI operating system, not just a good model: sensor streaming, memory management, fail-safes, and reliable updates all matter when “bricking a car is very expensive.” The big unlock for physical AI is not just smarter models, but making those models small, efficient, and reliable enough to survive the real world. (Time 0:43:50)