AI's Power Problem: The Next Big Challenge

AI has come a long way in recent years, but one major hurdle remains: power consumption. As AI models become increasingly complex, the energy required to run them is growing exponentially. In an effort to address this issue, Unconventional AI is working on a new kind of oscillator-based computer architecture that promises to reduce power use by as much as 1,000 times.

A New Kind of Computer

Led by Naveen Rao, formerly the head of AI at Databricks, Unconventional AI is developing a new kind of computer that uses oscillators instead of traditional chips. This approach has the potential to revolutionize the way we think about computing, and Rao believes it's the key to unlocking the next level of AI performance.

The output from the new Un-0 model is similar to that of image-generation models like Stable Diffusion or OpenAI's GPT Image 1. The impressive part is how it arrives at that performance. The model is built on an oscillator-based architecture that is completely different from the chips that power conventional computing and traditional LLMs.

• The advantages of the oscillator-based computing are complex, but Rao believes it will ultimately reduce power use by as much as 1000 times.
• Much of the infrastructure to get there is still being built. The current version of Un-0 runs on a software simulation of Unconventional's oscillator chips, but the company plans to release schematics for an actual chip soon.
• From there, the plan is to build an entire inference stack from the ground up, with Unconventional AI eventually supplying compute capacity just like any other provider.

A Bold Ambition

Unconventional AI's goal is nothing short of stunning: to build a new kind of system composed of their chips, which will run AI models at 1/1000 of the power consumption of traditional systems. It's a bold ambition, particularly for a company that still counts less than 50 employees. But given the scale of the AI buildout and the anticipated cost of meeting the growing demand for inference, it may be one of the few efforts to meet the scale of the problem.

AI scaling is hard because of energy. It's going to be the fundamental limit in the next few years. You just can't go past it. It’s going to be an energy limited problem, at the end of the day.