The UK says it wants to lead the artificial-intelligence (AI) revolution. But the infrastructure needed to power it is not being built.
A visit to Silicon Valley, where I am writing this from, makes the contrast clear. At NVIDIA’s annual GTC conference in San Jose – where much of the global AI ecosystem gathers – the conversation is about scale: the next generation of large language models (LLMs), the explosive demand for AI compute, and the vast data-centre capacity required to support it. Against that backdrop, the UK’s ambitions look increasingly detached from reality.
Over the last 12 months, British ministers and civil servants have repeatedly suggested the country could see 10–12GW of AI-related data-centre demand within the next five to seven years. Within the industry, and across the exhibition floor here in San Jose, that figure is widely viewed as implausible, and even more worryingly, completely unfounded.
At times, the sovereign AI optimism from our government resembles The Emperor’s New Clothes. If ministers and departments repeat often enough that an AI windfall is coming, perhaps we will all eventually believe it. The more realistic outlook is closer to 2GW of additional cloud and inference capacity over the same period.
That gap matters. It is not simply measured in megawatts. It represents as much as £150bn in potential data centre construction investment, before accounting for the far larger ecosystem that grows around digital infrastructure: jobs, construction, supply chains, software companies, AI startups, innovation and inward investment.
In the global AI race, power increasingly lies with the cloud and neocloud organisations that build and operate LLMs. Those companies need three things above all else: vast computational GPU capacity, predictable and welcoming AI regulation, and most crucially, competitive energy costs.
Right now, the UK is struggling to offer one, let alone all three, and there doesn’t seem to be much light right now at the end of a very dark tunnel.
Industrial electricity prices in the UK remain among the highest in the developed world (and this is before we see the true impact of the Iran conflict on a country heavily reliant on imported gas). The government’s approach to AI copyright reform remains both uncertain and snail-paced. And the broader policy framework designed to attract large-scale AI infrastructure still feels tentative and commercially confused.
Taken together, these signals matter. Investors deciding where to place tens of billions in infrastructure look for stability, scale, conviction and speed. On those measures, the UK is falling behind.
The government’s proposed AI Growth Zones further illustrate the problem. In theory they are designed to concentrate infrastructure investment and accelerate development, which we at Kao Data both welcome and embrace. In practice, however, many appear more conceptual than concrete.
Lanarkshire offers an exception, thanks to projects involving DataVita and the US neocloud, CoreWeave, this AIGZ designation supports a UK data centre business – DataVita housing Coreweave’s GPU platfrom. Yet these developments largely pre-date the growth zone designation and cannot easily be credited to it. Elsewhere – at sites such as Culham, Anglesey and the North East – there is little visible evidence of imminent data-centre construction.
Even regions with established digital infrastructure, such as South Wales, owe their position largely to earlier investments by companies such as Microsoft and Vantage rather than to recent policy initiatives.
Last September – timed conveniently alongside President Trump’s visit to the UK – the government proudly (and quickly… ) announced a series of initiatives framed as evidence of the UK’s AI leadership. Yet much of what was presented appeared to be investments that the cloud companies were due to make and have done so in other European countries. Some were also previously announced initiatives rebranded within a new AI narrative, rather than genuinely new government-led programmes.
Meanwhile, hyperscale cloud providers are building aggressively across continental Europe. Their strategy is straightforward: integrate AI inference workloads into existing cloud platforms – deepening regional ecosystems and reinforcing their global scale.
Recent industry data from the likes of CBRE suggests markets such as Paris and Frankfurt are now pulling ahead of London in new data centre deployment. The reality is the AI race is on, and the UK is ceasing to keep pace. If that trend continues, the UK risks drifting into an uncomfortable position in the AI economy -not as a producer of foundational technologies, but primarily as a backwater user of them. We will most certainly be ‘taking AI’, and not ‘making AI’.
None of this is inevitable. The UK retains many advantages: a strong research base, deep capital markets and a world-class technology sector. But capturing the infrastructure layer of the AI economy will require more deliberate and faster action.
That means lowering energy costs for digital infrastructure, investing in the grid and enabling infrastructure around potential AI hubs (and not just published AIGZs), and quickly creating a regulatory framework that actively encourages LLM development and AI training workloads.
Without those changes, the UK may discover that announcing an AI strategy is far easier than building an AI economy. After all, from where I sit here in San Jose, the real strategic choice facing the UK is simple:
Be a nation that builds the machines and produces AI – or become one that simply buys the AI from others.
