Our first investments
On 16th April we announce our first investment and the direction we’re setting.
Callosum
Our first investment is Callosum.
Callosum is building systems software for a world where compute is no longer uniform. Instead of assuming everything runs on a single class of hardware, Callosum treats modern AI infrastructure as it actually is: fragmented, heterogeneous, and increasingly complex.
Their platform orchestrates different chip architectures in real time – co-optimising models, workloads, and silicon together. The result is step-function improvements in both performance and cost.
Founded in Cambridge by Danyal Akarca and Jascha Achterberg, and headquartered in London, the company sits at the centre of a growing UK advantage in next-generation compute.
As our first investment, Callosum reflects exactly what we want to back: deep technical infrastructure that shapes how AI is computed globally, built and anchored in the UK.
Deploying national compute
Alongside our first investment, we’ve made our first large-scale compute allocations.
Through the AI Research Resource (AIRR), we’ve deployed over 3 million GPU hours – worth an estimated £14 million – to six frontier AI companies in the UK.
This is not a grant programme. It’s targeted infrastructure deployment. We focus compute where the UK has real structural advantages, progress is bottlenecked by access to large-scale infrastructure, and the upside is nationally strategic.
The companies
Prima Mente is building biological foundation models to decode the language of biology – from DNA to gene expression – to better understand diseases like Alzheimer’s and Parkinson’s. Working closely with Oxford, Imperial, and Edinburgh, they sit at the intersection of two areas where the UK is globally strong: AI and life sciences.
Doubleword is focused on inference infrastructure – helping companies actually run models in production. Their work in optimisation and compression makes it economically viable to deploy AI at scale, not just train it.
Cosine is building sovereign, air-gapped AI systems for defence and regulated industries. Their models run entirely within customer-controlled environments, with no external dependencies – critical for sectors where security and assurance are non-negotiable.
Cursive is developing agents that improve continuously from real-world use. This is one of the hardest and most important problems in AI today: systems that don’t just perform, but learn over time in deployment.
Odyssey is building world models that learn across modalities – vision, language, sound – to simulate and understand environments more like humans do. This unlocks applications across robotics, defence, and simulation.
Twig Bio is building CANOPY, a foundation model for engineering biology and biomanufacturing. The goal is to move from research to industrial-scale biological design – turning UK strength in synthetic biology into production capability.
Want to join our portfolio of high-impact AI startups? Get in touch.





