AI and hardware company Normal Computing UK was picked as one of 12 teams awarded funding from the Advanced Research + Invention Agency (ARIA) Scaling Compute Programme. This program, backed by £50 million in funding, aims to reduce AI hardware costs by 1000x while diversifying the semiconductor supply chain.

NEW YORK--(BUSINESS WIRE)--Normal Computing, a deep tech AI startup founded by former Google Brain and Alphabet X engineers to develop full-stack applications with enterprise reliability, today unveiled the world’s first thermodynamic computer. Normal’s team conducted the first-ever thermodynamic AI experiment using the prototype hardware to add reliability and controls to the outputs of a neural network – work that could one day help eliminate hallucinations in AI models, and enable AI agents that can reason about the world, yet are controllable and safe. AI applications, like those powered by generative AI and large language models, require massive resources, and today’s computers may not be powerful enough to unlock the full scope of applications. However, the energy required for today's advanced computers will only become a bigger problem as AI models grow in size, with energy consumption already a major issue for today’s Graphical Processing Units (GPUs). Furthermore, even cutting-edge generative AI solutions can be unreliable and unusable in mission-critical applications. Properly accounting for uncertainty using probabilistic AI methods may be essential for AI agents to plan, reason, and have common sense