Naveen Rao: 4D Computing, AI's Energy Wall & Beating Biology

Video summary

Naveen Rao unveils Unconventional AI’s 4D computer, targeting 1,000x efficiency and a two-year product launch.

Naveen Rao, co-founder and CEO of Unconventional AI, argues that AI’s energy demands are approaching a wall: Google processes 3.2 quadrillion tokens monthly, consuming an estimated 12 gigawatts, while energy represents about 50% of token-serving costs. His company is replacing conventional von Neumann architecture with a dynamical computer that combines memory and compute, using physical oscillators, sparsity, and time plus 3D chip stacking—what Rao calls “4D computing.” A prototype built in five months generated images at roughly 500 nanojoules each, versus millijoules for GPUs. Rao targets 1,000x greater power efficiency, expects a full rack-scale data-center product within two years, and says existing models can be ported at the model layer. The long-term ambition is to beat biology and enable localized AI and billions of robots.

Chapters

1. 0:00Welcome Naveen Rao: From Nirvana Systems to Unconventional AI
2. 4:45Energy Wall: Google’s 3.2 Quadrillion Tokens Require 12 Gigawatts
3. 10:24Dynamical Systems: Unconventional AI’s Uno Image Model and Sparse Connections
4. 14:584D Computing: A Physical Dynamical Computer Uses 500 Nanojoules per Image
5. 19:40Product Path: Unconventional AI Plans a Full Data-Center Rack Within Two Years