This Startup Wants to Make AI Chips 1,000 Times More Efficient

Most AI chip companies are trying to squeeze more performance out of familiar architectures. Unconventional AI is taking a more radical approach: rethink the way computation itself happens.

The young hardware startup is acquiring Israeli chipmaker NeuReality in a cash-and-stock deal designed to speed up product development. Unconventional AI says its long-term goal is to build AI chips that are up to 1,000 times more energy efficient than today's systems.

A different way to think about computing

The company is exploring what founder Naveen Rao describes as a dynamical-systems approach, where simple computing elements interact and evolve toward a result rather than executing a rigid sequence of instructions.

That idea is partly inspired by biological systems, which perform complex tasks with a fraction of the energy consumed by modern data centers.

Why NeuReality fits

NeuReality brings an experienced engineering team and several years of advanced AI chip development. About 40 employees are expected to join Unconventional AI, giving the startup more experience in turning research prototypes into manufacturable products.

Rao said the company had expected a much longer research timeline, but AI-assisted development has accelerated progress. It has already built one test chip and is nearing completion of another.

Efficiency may become the next AI battleground

The AI industry has focused heavily on performance, but energy is becoming just as important. Data centers are consuming more electricity, governments are worried about grid capacity, and chipmakers are under pressure to deliver more compute per watt.

If Unconventional AI can deliver even part of the efficiency improvement it is targeting, the impact could extend far beyond one startup. It could change the economics of training and running advanced AI systems.