Callosum is a UK AI-infrastructure startup challenging what its founders call the 'monoculture' of AI compute — the industry's heavy dependence on running ever-larger models across banks of identical Nvidia GPUs. Its software orchestrates AI workloads across a heterogeneous mix of chip types, intelligently distributing tasks to whatever silicon is best suited and available: Nvidia GPUs, AMD processors, Amazon Web Services' custom Trainium and Inferentia chips, or newer designs from startups like Cerebras and SambaNova. The goal is to give AI teams flexibility, resilience, and cost efficiency in an increasingly diverse and supply-constrained hardware landscape.
The company is named after the corpus callosum — the bundle of fibers connecting the brain's hemispheres — reflecting its founders' neuroscience roots and its mission of connecting disparate compute. Callosum was co-founded by Danyal Akarca and Jascha Achterberg, who met during their PhDs at the University of Cambridge around 2019, working at the intersection of the brain, computing, and AI. Their research, some conducted in collaboration with Google DeepMind, has been published in several Nature journals, giving the company a distinctive scientific foundation for thinking about how to map computation onto varied hardware substrates.
In February 2026, Callosum emerged from stealth with $10.25 million in funding. The round came from European VC firm Plural and ARIA — the UK government's Advanced Research and Invention Agency, charged with backing high-risk, high-reward science — along with angel investors including Charlie Songhurst, Stan Boland, and John Lazar. ARIA's involvement signals that Callosum's approach is viewed as a frontier research bet with potential strategic importance to the UK's compute independence.
Callosum addresses a real and growing problem: as alternative AI chips proliferate and GPU supply tightens, organizations want to avoid lock-in to any single vendor, but writing and optimizing software for many different accelerators is hard. By abstracting away hardware differences and scheduling workloads across a mixed fleet, Callosum aims to make heterogeneous AI compute practical — potentially lowering costs and improving access for teams that can't secure enough of a single chip type.
As an early-stage company with a modest seed round relative to the chip giants it complements, Callosum's path depends on proving that its orchestration delivers meaningful efficiency and reliability gains in production. Its brain-inspired framing and strong scientific pedigree differentiate it, but execution in the demanding world of AI infrastructure software will determine whether the vision of true cross-chip portability becomes reality.