The UK government has launched the Sovereign AI Unit, a £500 million initiative designed to back homegrown artificial intelligence companies from their earliest stages. With equity investment, access to world-class supercomputing infrastructure, and fast-track visa support on offer, the programme marks a significant shift in how the state engages with the technology sector. Seven startups have already been selected, working across fields from drug discovery to national security. The initiative is built to operate with the speed and conviction of a venture capital firm, rather than the measured pace typically associated with public sector programmes.
The unit will provide direct equity investment in the country’s most promising AI startups, alongside a package of practical support that includes access to some of Britain’s largest supercomputers, fast-track visa decisions, R&D funding, and routes into early government procurement. The stated ambition is clear: to ensure that world-class AI companies are founded, scaled, and retained within Britain, rather than losing their best ideas and talent to overseas competitors.
Technology Secretary Liz Kendall, announcing the programme at the London headquarters of Wayve, a self-driving technology company that grew out of research at the University of Cambridge, framed the effort as a matter of both economic strategy and national security. The UK, she said, must be an AI maker rather than an AI taker.
“We believe in Britain and we are betting on Britain. We are backing our brilliant innovators and entrepreneurs so we seize the benefits of AI to reshape Britain for the benefit of all.”
Liz Kendall, Technology Secretary, UK Government
The founding cohort
Seven startups have been selected as the initial beneficiaries of the Sovereign AI Unit. The first to receive direct equity investment is Callosum, a London-based infrastructure company founded by Cambridge PhD graduates Danyal Akarca and Jascha Achterberg. The company is building systems-level software designed to allow diverse chip architectures to work together, optimising models and silicon in real time. As the demand for AI inference grows and the hardware landscape fragments beyond conventional GPU architectures, this kind of orchestration layer is expected to become increasingly important.
Six further startups will receive access to the AI Research Resource (AIRR), the UK’s national supercomputing network, with each allocation providing the processing power required to train advanced AI models, run complex simulations, and test ideas at scale. The unit has also agreed rights of first refusal on future investment in several of these companies, creating a pathway from early compute support to longer-term financial backing.
From biology to national security
The six AIRR recipients cover a wide range of strategic fields. Prima Mente is using AI to analyse biological systems at the molecular level, building foundation models aimed at improving the understanding and treatment of brain diseases such as Alzheimer’s and Parkinson’s. The company works in active collaboration with Oxford, Imperial College London, and the University of Edinburgh.
Cosine is developing advanced AI models and coding agents purpose-built for defence and national security applications, where foreign-built AI platforms are not a viable option. Its platform operates entirely within a customer’s own infrastructure, with no external dependencies. Cursive, founded by senior alumni of DeepMind, is building AI agents capable of continuous learning from real-world deployment, a frontier currently pursued by only a small number of research organisations globally.
Twig Bio is developing a foundation model called CANOPY for AI-driven strain design in engineering biology, with the goal of making biomanufacturing viable for a broader range of sustainable ingredients. Doubleword provides the infrastructure needed for AI inference, enabling organisations to adopt AI faster and at lower cost, while allowing regulated and government users to run and audit models within their own secure environments. Odyssey, founded by British researchers Oliver Cameron and Jeff Hawke, is building world models that learn from multiple types of input, with applications across defence, autonomous systems, and simulation.
“AI as a technology could be transformational for both our wealth and security. Britain has the foundations to be a global AI leader in many fields, with a unique and enviable mix of talent, capital, and infrastructure.”
James Wise, Chair, Sovereign AI Unit
The full package
Beyond funding, companies backed by the Sovereign AI Unit will receive access to up to one million GPU hours per startup through the AIRR network, giving them the computing power required to train state-of-the-art models without the prohibitive hardware costs that typically constrain early-stage companies. Visa decisions for international talent will be processed within a single working day, with an initial allocation of ten cost-free visas available to each investment recipient.
The unit will also offer hands-on support navigating access to government data, early procurement opportunities, independent product validation, and engagement with new regulatory frameworks. For startups working in fields where government is both a potential customer and a gatekeeper, this kind of direct access carries significant practical value.
As part of its broader £282 million commitment to support cutting-edge AI research and development, the unit has launched its first funding call to create new datasets and related assets designed to help firms move faster and build more effectively within the UK. A further thirty companies are currently in discussions with the unit regarding potential AIRR access, with further allocations expected over the course of the year.
Built to move fast
What distinguishes the Sovereign AI Unit from previous government-backed technology initiatives is the deliberate adoption of private-sector operating principles. Rather than issuing grants through lengthy application processes, the unit is designed to move at the pace of the industry it is supporting, making investment decisions quickly and working alongside founders rather than around them.
Chancellor of the Exchequer Rachel Reeves described a thriving domestic AI sector as one of three priority choices for the economy, positioning support for strategic AI companies as central to the government’s growth agenda. The unit will work in parallel with the government’s Global Talent Taskforce to attract leading researchers to the UK, and will begin a tour of cities across the country in May to ensure the programme’s benefits reach beyond the London tech corridor.
“We don’t have a talent problem in the UK, we have a scale problem. The next wave of AI winners will come from countries that don’t just invent, but back their builders end-to-end.”
Alex DePledge, Entrepreneurship Advisor to the Chancellor, UK Government
Backing builders to stay
The unit’s investment logic is as much about retention as it is about discovery. For years, a familiar pattern has played out across the British technology sector: companies begin in the UK, attract early-stage funding, and then relocate or sell to larger foreign players as they grow. The Sovereign AI Unit is designed to interrupt that cycle by providing the conditions and capital required to support companies through their scaling phase without forcing a choice between ambition and geography.
For Callosum’s founder Danyal Akarca, the UK’s depth of talent across universities and research institutions such as DeepMind makes it a natural home for infrastructure work that sits at the centre of how the next generation of AI systems will be built and operated. For Cosine, the AIRR allocation is significant precisely because it allows the company’s own models to be trained on sovereign infrastructure, removing a dependency on foreign platforms in sectors where control over AI capability is non-negotiable.
The Sovereign AI Unit will continue to assess applications on a rolling basis, with compute allocations worth tens of millions of pounds expected to be distributed to British startups across the year. Whether this pace of deployment proves sufficient to keep Britain competitive with larger national AI programmes elsewhere remains to be seen, but the intent to act with urgency is, at least, clearly stated.