The UK wants ‘AI sovereignty’ to cut its reliance on foreign tech – here’s how that can be achieved

The UK government wants to grow the country’s native AI industry to reduce its dependence on systems made by foreign companies. In the Budget on October 28, Chancellor John Healey announced more money – £60 million – for the national Sovereign AI Fund to support promising UK AI companies. The government launched the fund in April 2026, committing £500 million at the time to boost the UK’s AI capability. Based in the Department for Science, Innovation and Technology (DSIT), it operates much like a government backed venture capital fund. But this unit offers more than money. Companies can gain access to some of the UK’s most powerful AI supercomputers, including Isambard-AI, alongside research support and opportunities to work with the government. The UK government initiatives are part of a wider global push towards sovereign AI. This term refers to countries strengthening their ability to develop and use the AI technologies they rely on while retaining greater control over them. There is concern about the risks of dependence on foreign suppliers – and the potential disruption to UK infrastructure if something were to happen to one of those suppliers. But what does sovereign AI mean in practice – and can the UK learn anything from what’s going on elsewhere in the world? For some, sovereignty means controlling everything needed to develop and run AI. For others, it means building domestic AI models or keeping data within national borders. A recent analysis published by the Carnegie Endowment for International Peace shows how complex this can be. AI depends on computer chips, data, electricity, skilled workers, internet connections, and cybersecurity. A country may control some of these while depending on foreign suppliers for others. In practice, four ideas tend to be bundled together. The first is ownership: who owns the chips, data centres, and models? The second is jurisdiction: whose laws apply and where is the data held? Sovereign AI featured in John Healey’s first budget as Chancellor. Tolga Akmen The third is capability: can a country build, adapt, and run AI itself? The fourth is resilience: can essential services continue if a supplier withdraws access? Many disagreements about sovereign AI concern which of these ideas matters most. The Center for a New American Security (CNAS), a US think tank, tracks government-backed sovereign AI projects. In January 2023, it recorded just one project. By mid-2026, that number had risen to 184 projects across 67 countries. These figures show how quickly interest in sovereign AI has grown globally. However, they count projects, not whether those projects have succeeded. Countries are making different choices about where to invest. India, for example, has focused on building domestic computing capacity, promoting the development of AI models and setting up national datasets for “training” AI models. Its IndiaAI Mission has deployed around 38,000 graphics processing units (GPUs) – specialised processors used to develop and run AI – to boost computing power for developers. India recently hosted a major AI summit. The country’s government has been boosting computing power to support homegrown AI startups. Rajat Gupta / EPA Images Other Indian initiatives include BharatGen AI, a multilingual AI model, and AIKosh, a national platform to provide startups and researchers with access to anonymised training data. According to the same report by CNAS, most countries focus on just one or two layers of the AI stack – the collection of technologies and infrastructure needed to develop and operate AI systems. Only a handful of jurisdictions, such as India, South Korea, the UK, and Taiwan, have operational programmes covering computing infrastructure, models, and data. Ghana and South Africa, meanwhile, have launched projects focused on expanding domestic AI computing capacity. Sovereign and also dependent? While some nations are making strides towards sovereign AI, almost every country relies on foreign AI technology to some extent. Rather than developing models from scratch, many adapt existing – or base – models developed abroad. According to CNAS, around three-fifths of the foreign base models identified in the projects it tracks come from the US. A 2026 report by researchers from the Center for Security and Emerging Technology (CSET) at Georgetown University, Washington DC, highlights the risks of foreign technology supply chains. Trade restrictions and geopolitical tensions can disrupt access, particularly to advanced computer chips, which are produced by a small number of specialised suppliers. But dependence is not the whole story. Isambard-AI: the UK’s most powerful AI supercomputer (University of Bristol / Bristol Centre for Supercomputing) Researchers at Stanford University in California have argued that sovereignty is better understood as managing AI dependencies rather than eliminating them. Their report underlines the fact that supporting domestic companies and adopting open-source technologies can widen the available choices. Diversifying foreign technology suppliers is another option. For example, the CNAS report identifies projects in Egypt, Singapore, Thailand, Uganda, and the UAE using Alibaba’s Qwen models, developed in China. Many Qwen models are open source and offer an alternative to those developed in Silicon Valley. Switzerland takes a different approach. Its Apertus model is a Swiss-developed large language model that openly shares information about its training data, code, and development process. In the Swiss canton of Ticino, it powers an in-house translation service for sensitive government documents without relying on commercial AI providers. The priority here is greater control over how the technology is used, rather than simply choosing the most powerful AI model. Less a switch, more a dial There are limits, however. As the Stanford researchers point out, commercial products marketed as “sovereign AI” may still rely on foreign providers. Even models that can be freely adapted still need chips, electricity, and skilled people. If sovereignty means owning every part of the AI technology stack, almost no country qualifies. But if it means maintaining control when circumstances change, the picture looks different. What matters is whether countries have meaningful choices. Can they switch suppliers without disrupting essential services? Can they enforce their own laws and adapt AI technologies to local languages, priorities, and needs? These questions reveal more about a country’s sovereignty than simply counting how many chips, data centres, or AI models it owns. Trade restrictions and geopolitical tensions can disrupt access to advanced computer chips. IM Imagery / Shutterstock Sovereignty then looks less like a switch and more like a dial. For the UK, the real test will be whether its Sovereign AI Unit gives the country greater freedom to choose and adapt AI technologies as the programme develops. “Sovereign AI” will probably continue to mean different things to different governments. But the label itself tells us little about how much control a country actually has.