Normal Computing raises $50M from Samsung to build thermodynamic AI chips targeting the data center energy crisis
Mar 27, 2026 with Faris Sbahi
Key Points
- Normal Computing raises $50M from Samsung to build thermodynamic AI chips that exploit noise inherent in diffusion models, targeting a potential 49-gigawatt data center power shortfall by 2028.
- OpenAI's shutdown of Sora after the model burned $15M daily against $2.1M in cumulative revenue exemplifies the economically broken workloads thermodynamic computing aims to fix.
- Normal Computing targets production readiness within five years, with 2030 as the internal delivery date for chips designed to solve the energy crisis before the industry hits a hard physical wall.
Summary
Normal Computing raised $50M led by Samsung in what the company calls an accelerator round. Founder Faris Sbahi came out of Google Brain and Google X and is betting that thermodynamic chips can solve a problem conventional silicon cannot: the coming data center energy crisis.
Thermodynamic chips are inherently noisy and approximate, which makes them poorly suited for deterministic tasks like arithmetic but well-suited for workloads like diffusion models, where noise is already part of the computation. The hardware physics maps directly onto the math, potentially cutting energy consumption dramatically.
Normal Computing targets 2030 for production. The company projects that by 2028, the industry faces a 49 gigawatt power shortfall. Something has to give whether that is new energy sources, data centers in space, or a genuine breakthrough in chip efficiency. Normal Computing is pursuing the last option.
The commercial urgency is acute. OpenAI shut down Sora this week after the video diffusion model cost roughly $15M per day to run against $2.1M in cumulative revenue. That kind of economically broken workload is exactly what thermodynamic computing is designed to fix. Whether Normal Computing reaches production readiness before the energy wall hits remains open, but the team puts the timeline at under five years.