Databricks Former AI Chief Wants to Cut AI’s Power Costs by 1,000 Times

Unconventional AI is developing a new computing architecture that could make AI inference far more energy efficient

As artificial intelligence systems become more powerful, their energy demands are becoming harder to ignore. Training large models already requires enormous computing resources, and running those models for everyday use may become an even bigger challenge.

Naveen Rao, the former head of AI at Databricks, believes the answer is not simply building bigger data centers. His new company, Unconventional AI, is trying to rebuild the way AI computing works from the ground up.

The company claims its approach could reduce the power needed for AI inference by as much as 1,000 times.

A New Approach to AI Computing

Unconventional AI is working on an oscillator based computing architecture. This is different from the traditional chips used to power most AI systems today.

Instead of relying on standard computing methods, the company is building a system designed specifically for AI workloads. The goal is to make inference faster, cheaper, and far more energy efficient.

Inference is the process that happens when an AI model responds to a user prompt, creates an image, writes text, or performs another task after training is complete. As more people use AI tools every day, inference costs are expected to become a major burden for the industry.

The First Test Model

Unconventional AI has released its first model, called Un 0. It is an image generation model built to show that the company’s architecture can support modern AI tasks.

The current version runs through a software simulation of the company’s planned hardware. This means Unconventional AI has not yet fully deployed its custom chips, but it is using the model to prove that the architecture can work.

The company says Un 0 can produce results similar to well known image generation systems. The bigger point, however, is not just the images themselves. It is the way the model reaches those results using a different computing design.

Why Energy Efficiency Matters

AI companies are spending heavily on chips, servers, cooling systems, and power supply. As demand grows, energy may become one of the biggest limits on how quickly AI can scale.

If every search, chatbot response, image, video, or agent task requires large amounts of electricity, the cost of running AI could rise sharply. That could make advanced AI harder to access and more expensive for businesses and consumers.

Unconventional AI is targeting this problem directly. By lowering power use for inference, the company hopes to make AI more sustainable and more affordable at scale.

From Simulation to Real Chips

The company’s next major step is to move beyond software simulation and toward real hardware. It plans to release schematics for its chip design and eventually build a complete inference system around its technology.

The long term vision is to provide computing capacity to customers, similar to how cloud providers and AI infrastructure companies sell access to processing power today.

If successful, customers would send prompts into Unconventional AI’s system and receive model outputs while using a fraction of the electricity required by conventional infrastructure.

A Small Company With a Large Ambition

Unconventional AI is still a relatively small startup, with fewer than 50 employees. Its goal, however, is extremely ambitious.

Rebuilding AI infrastructure is not easy. The company must prove that its architecture can work outside simulations, support real customer workloads, and compete with the powerful chips already used across the AI industry.

It also needs to show that its efficiency claims can hold up at scale. A promising demonstration is only the first step. Large customers will want reliability, speed, compatibility, and clear cost savings before changing how they run AI systems.

A Possible Shift in the AI Race

The AI industry has often focused on bigger models and more computing power. Unconventional AI is taking a different path by focusing on how to reduce the energy cost behind those systems.

If the company succeeds, it could change the economics of AI. Lower power use could make advanced models cheaper to run, easier to scale, and less dependent on massive energy infrastructure.

The challenge is still enormous, but the idea reflects a growing reality in the AI world. The future of artificial intelligence may not depend only on smarter models. It may also depend on finding better ways to power them.