Summary
Naveen Rao, co-founder and CEO of chip startup Unconventional AI and former founder of Nervana Systems and MosaicML, delivers a keynote at the All-In Podcast summit on why AI hardware needs to be rethought from first principles. Rao argues that today’s computers were designed to be fast, not energy-efficient, and that decades of digital abstraction — turning continuous physics into ones and zeros — is fundamentally lossy and wasteful of power.
He lays out Unconventional AI’s approach to closing what he calls AI’s ‘energy wall,’ describing a stack that runs from theoretical neuroscientists and mathematicians down to physical circuit designers, all working toward a goal of 1000x power efficiency gains — a timeline he says has accelerated from five years to roughly three and a half. Rao draws on biological analogies, including neurons, bird flocking, and ant colonies, to explain how simple physical rules can produce emergent intelligence without traditional linear algebra or floating-point math.
The talk also touches on his career path from Intel’s AI group to Databricks (after MosaicML’s acquisition) and his current bet that new computing substrates, not just bigger models, will be necessary for AI to keep scaling. It’s a technical but accessible look at the hardware side of the AI industry for viewers curious about what comes after GPUs.
📺 Source: All-In Podcast · Published September 21, 2026
🏷️ Format: Keynote Launch
