The London-based startup Callosum Technologies, founded by Cambridge-trained neuroscientists Danyal Akarca and Jascha Achterberg, has raised more than $110M in cumulative funding to build orchestration software that breaks AI workloads into subtasks and assigns each one to the best possible combination of chips and models.
From stealth to $100M seed round
Callosum emerged from stealth on February 26, 2026, backed by $10.25M in pre-seed funding. By August 2026, Callosum announced a $100M seed round led by Atomico, with participation from Plural, DCVC, and the UK’s Sovereign AI Fund, putting it among the largest seed rounds ever raised in Europe.
On top of the venture funding, Callosum secured a $2.9M grant from ARIA, the UK’s Advanced Research and Invention Agency, earmarked for heterogeneous computing research.
The neuroscience angle
The corpus callosum is the band of nerve fibers connecting the brain’s two hemispheres, allowing specialized regions to coordinate on complex tasks. Callosum’s software applies the same logic to AI inference: it dissects incoming workloads, identifies the computational characteristics of each subtask, and dynamically routes them across a diverse hardware stack.
That stack can include Nvidia GPUs, AMD processors, AWS Trainium and Inferentia chips, and other emerging accelerators. The company claims this approach delivers up to 2x accuracy improvements, 7x speed gains, and 4x cost savings compared to traditional homogeneous GPU setups on complex AI tasks.
Why inference matters more than ever
Deloitte has estimated that roughly two-thirds of AI compute will be dedicated to inference by 2026. Callosum’s software supports ultra-low-latency tasks on cloud platforms like AWS and optimizes across varied hardware, making it particularly suited for complex, multi-step agentic AI workflows.