Naveen Rao, founder of Nirvana Systems and CEO of Unconventional AI, argues that AI faces an imminent energy wall, with Google alone consuming ~12 GW for inference. He introduces a radical new computer architecture—a 'dynamical computer' that combines compute and memory—inspired by the brain's efficiency. Rao reveals the first physical prototype, achieving 500 nanojoules per image and aiming for 1,000x power efficiency within 3.5 years.
Summarized by Podsumo
Google alone uses ~12 GW/month for AI, and the world is on track to run out of data center energy in ~3 years.
The human brain runs on 20 watts; the new chip achieves 500 nanojoules per image, thousands of times more efficient than GPUs.
Rao revealed the first physical dynamical computer, built in 5 months, that generates images using oscillators instead of traditional linear algebra.
Sparsity—removing connections in neural networks—actually improves scalability and performance in this new architecture.
The company aims to hit the thermodynamic limits of computation within 3.5 years, creating a system 1,000x more power efficient than current hardware.
"The way we think about data centers has shifted... Today it's about energy. First, you think about energy."
"We want to monetize that [watt] a thousand X better than existing hardware."
"The overarching goal of this company is to beat biology."