Heavy Strategy HS126: AI Everything, AI Everywhere, AI All At Once
9 snips
Mar 10, 2026 A wide-ranging take on chip vendors pitching AI across cloud, desktop, factories and enterprise software. Debate over whether everything should be rewritten as AI and the heavy compute and energy costs that implies. Concerns about agentic UIs, security and hallucinations in physical systems. Discussion of simulation as a new infrastructure pillar, digital twins for industry, and when IT should partner with chip makers.
AI Snips
Chapters
Transcript
Episode notes
AI As A Unifying Layer Looks Impractical Today
- NVIDIA framed AI as a unifying layer that turns applications into things you train rather than write.
- John Burke warned this vision demands enormous compute and energy and is currently infeasible at broad scale.
Agentic AI Interfaces Need Stronger Assurance
- Jensen promoted agentic AI as the future user interface but barely addressed security or assurance.
- John Burke and Johna Till Johnson flagged hallucination and lack of 'I don't know' behavior as dangerous for agentic systems.
Simulation Pillar Creates New Data Center Demands
- NVIDIA proposes a third AI pillar: simulation, in addition to training and inference, to generate data for models.
- Hosts stressed simulation must live adjacent to inference with high-speed I/O and will add major data center and networking needs.
