Callosum is a London-based AI infrastructure company founded in 2025 by Cambridge-trained neuroscientists Danyal Akarca and Jascha Achterberg. It builds software that breaks AI workloads into sub-tasks, matches each sub-task to the model size it needs, and routes it to the chip that runs it most efficiently. The platform integrates at the compiler layer, so supporting a new accelerator is a software task, and performance data feeds back to improve routing over time. Its product, Tailored Inference, is offered as an enterprise API family for developers and enterprises with large inference bills.

The company's thesis is that AI workloads are heterogeneous and should run on a mix of models and silicon rather than on identical chips and a single model. Its technology is designed to work with GPUs from Nvidia and AMD, Google TPUs, AWS Trainium, Cerebras wafer-scale systems, SambaNova, and emerging photonic and neuromorphic architectures. Callosum has announced a flagship partnership with Cerebras to integrate its WSE accelerators into Tailored Inference, and it has also announced partnerships with Rebellions and other chip and infrastructure providers.

Callosum emerged from stealth in February 2026 with a $10.25M pre-seed round led by Plural, with support from ARIA. In April 2026 it became the first investment of the UK Sovereign AI Fund. In August 2026 it closed a $100M seed round led by Atomico, with participation from Plural, DCVC, the UK Sovereign AI Fund, Dunamu & Partners and other investors and angels. The company reports a team of around 40 to 45 people across the UK, the US, India and Sweden, and states that the new funding will be used to scale its systems software and expand its engineering team.