The idea that every country should have its own artificial-intelligence infrastructure—data centres, chips and models trained on local languages and data. Jensen Huang, Nvidia's boss, began advocating the concept in late 2023. At least 20 countries are now pursuing sovereign AI.
The European Commission unveiled plans for a €20bn fund for up to five "AI gigafactories." France, Germany, Saudi Arabia, South Korea and the UAE have all been involved in deals for local AI infrastructure. Saudi Arabia expects to purchase "several hundred thousand" of Nvidia's top-end processors over five years; the UAE intends to import half a million annually. Jefferies, an investment bank, estimates sovereign initiatives could generate $200bn in cumulative revenue for Nvidia; Nvidia itself reckons spending could reach $1trn.
South-East Asia is becoming a vital arena in the global tech competition between America and China. Singapore, Malaysia, Thailand and Indonesia together host nearly 2GW in data-centre capacity, equivalent to London and Frankfurt combined. Alibaba has partnered with Tencent and GoTo to train Indonesians in cloud computing and AI; Tencent alone plans to invest $500m in Indonesia by 2030. Amazon and Microsoft make up 60% of the cloud "infrastructure as a service" market in the region. Alphabet is investing $1bn in Thailand. Huawei has begun selling its Ascend AI chips as an alternative to Nvidia's banned semiconductors—evidence that China is building a rival technological stack rather than merely evading American export controls.
Britain's newest and biggest supercomputer, Isambard AI, is housed in a car park on the outskirts of Bristol. Commissioned in August 2023 by then-prime minister Rishi Sunak, it was switched on in July 2024 -- remarkably fast, since a supercomputer typically takes four to five years to build. It contains 5,448 Nvidia chips and has more computing muscle than all other British supercomputers combined. It is the world's 11th-fastest supercomputer; Finland, Germany, Italy, Japan and Switzerland all have more powerful resources.
Isambard is not big enough to train the largest language models but enables research breakthroughs. Around 80% of its 100-plus current projects are in life sciences and health, from skin-cancer detection to drug design. It is named after Isambard Kingdom Brunel, a Victorian pioneer. The machine is directed by Simon McIntosh-Smith, a professor at Bristol University and director of the Bristol Centre for Supercomputing.
Keir Starmer has pledged £2bn for AI investment, including £750m for a supercomputer in Edinburgh -- a rounding error next to the $320bn America's biggest tech firms will spend in a single year on AI infrastructure. Britain is deficient in the resources to match other middle powers in building data centres: the UAE has oil wealth and abundant solar power, Canada has cheap energy and land, whereas Britain has some of the most expensive electricity in Europe.
Britain's best chance lies in nurturing AI talent. It is a world leader in machine-learning research. DeepMind, Google's AI arm, and Wayve, a promising self-driving-software firm, are based in London's King's Cross. Matt Clifford, until recently the government's AI czar, argues that one of Britain's greatest opportunities could be using AI to reimagine public services such as the health-care system.
Governments justify sovereign AI on several grounds: catching up to America, ensuring AI models incorporate local languages and values, controlling domestic data (particularly health care), and widening access for smaller companies. Denmark's Gefion supercomputer is used for drug discovery and weather forecasting. Critics argue these projects risk creating "more like a palace than a factory" and that America's cloud giants could provide sovereign clouds more efficiently.
Bowie's Theorem: If an experiment works, you must be using the wrong equipment.