Callosum Raises $100M To Cut AI Costs By Routing Workloads To Specialized Chips

Callosum, a startup founded by Cambridge neuroscientists, has secured $100 million in funding to revolutionize the way AI tasks are processed. The company aims to cut AI costs by routing workloads to specialized chips, a strategy inspired by the human brain's use of small, specialized circuits for specific jobs.

The $100 million round was led by Atomico, with participation from Plural and DCVC, and marks the first investment from the UK's new Sovereign AI Fund. This funding comes at a critical time as AI moves from training massive models to deploying them at scale, with companies discovering that running these models is where the real money is burned.

Most AI companies currently run every task on expensive, general-purpose Nvidia GPUs, regardless of the task's requirements. This brute-force approach is driving up budgets fast, with inference often taking up over 50% of revenue for AI-native companies. Callosum's solution is to stop treating all compute the same, just like the human brain.

The company's software acts as a smart routing layer between applications and hardware, breaking down complex AI tasks and automatically assigning each piece to the cheapest, most efficient model and chip available. This approach has already shown results, with early deployments reporting 70% reductions in cost and 4x speed gains when paired with specialized chipmakers like Axelera, Cerebras, and Rebellions.

The AI market is shifting focus from training foundational models to deploying them cheaply. Callosum is positioning itself at the center of this shift, aiming to change how software talks to physical microchips. With backing from Atomico and the UK government, the company is well-positioned to be a key player in the next major hardware battleground in AI compute.