For two decades, London held an effectively unchallenged position as the centre of the UK’s digital infrastructure. The cluster of data centres around Slough, West London, and the M4 corridor became the default home for the country’s financial systems, cloud services, and the underlying compute behind almost every consumer app used in Britain. The reason was straightforward: London had the fibre, the customers, and the connectivity to make latency-sensitive workloads worth the premium.
That equation has changed. The constraint that now defines where data centres can be built in and around the capital is not connectivity or property cost. It is electricity. And the AI boom has turned what was a long-term planning problem into an immediate commercial one.
The grid has run out of headroom
In July 2024, the Greater London Authority confirmed that parts of West London cannot accept new electricity connections of any meaningful scale until at least 2035. The areas affected are precisely the ones where most of the UK’s hyperscale and colocation capacity is concentrated. National Grid, the GLA, and the affected London boroughs have all acknowledged that the demand for new connections far exceeds what the regional grid can deliver in the near term.
For developers planning standard commercial property, that is an inconvenience. For AI infrastructure, it is a hard stop. A modern GPU rack used for AI training can draw 60 to 120 kilowatts continuously. A single hyperscale AI cluster can pull tens of megawatts. New AI projects requesting connections in West London today are routinely being told the power they need will not be available until well into the 2030s.
The result is a structural shift in how operators are thinking about location. AI infrastructure is no longer flowing automatically to London. It is being redirected to wherever grid headroom is available now.
The numbers behind the squeeze
The scale of the problem is set out in the UK government’s research briefings on data centre infrastructure. The UK currently operates around 1.6 gigawatts of national colocation capacity, with industry forecasts placing demand between 3.3 and 6.3 gigawatts by 2030. Data centres already consume roughly 2.5% of UK electricity, a figure expected to roughly quadruple within the decade as AI workloads scale.
The CBRE 2026 UK Data Centres Outlook reports that take-up has exceeded new supply for four consecutive years. New supply has been pushed outwards from the historic London cluster, into the Thames Valley, the M4 corridor, the Midlands, and increasingly the North West. Pricing reflects this. Power capacity, once a minor line item in colocation contracts, is now the single largest commercial variable in many deals.
What this means for London’s tech economy
The economic question for London is not whether the city will lose its data centre cluster outright. It will not. The latency-sensitive workloads that justified building in West London in the first place are still there. Financial trading, cloud peering, content delivery, and large-scale enterprise applications all retain a measurable advantage from being inside the cluster.
The question is what happens to the next generation of compute-heavy workloads, the ones being built now around large language models, generative AI, and high-performance computing. Those workloads need power more than they need proximity. And the operators building them out are no longer assuming London is the default answer.
The pattern emerging across the industry is split deployment. Latency-sensitive front-end systems remain in the capital cluster. The compute-heavy training and inference workloads sit further out, in regional hubs where grid capacity is actually available and where land, power, and cooling can be sourced together at scale.
Where the capital’s overflow is going
Several regional clusters are now absorbing what would previously have defaulted to London. The Thames Valley west of the capital still holds the closest combination of latency and modest available power. The M4 corridor extending through Reading, Newbury, and Swindon has become one of the most active corridors for new build-out in the country, with the power picture meaningfully better than central London and connectivity into the capital that supports most enterprise applications.
A Hayes project from UK data centre colocation operator Carbon-Z illustrates the trend. Based at The Shipping Building in UB3, the site provides 10.7 megavolt-amperes of contracted grid capacity supporting a UK data centre cloud operator, with the facility built specifically for high-density compute and currently moving through a twelve to eighteen-month construction and commissioning timeline. The decision to develop in Hayes rather than further inside the West London cluster reflects exactly the pattern now playing out across the sector: viable power capacity, accessible enough to London to retain its commercial value, and engineered from the ground up for the densities AI workloads require.
Further north, Manchester and the wider North West have become the focus of significant hyperscale and enterprise investment, partly because the regional grid still has meaningful headroom and partly because planning timelines are shorter than in the South East. Government policy has formally encouraged this shift through the introduction of AI Growth Zones, announced alongside the recognition of data centres as Critical National Infrastructure in September 2024.
The longer-term picture
What is happening now is not the end of London as a tech hub. It is the end of a specific assumption that the capital can absorb every kind of digital workload indefinitely.
The buildings, the talent, the fibre, the financial services that need to sit inside the cluster, all of that remains. What is moving is the new wave of compute-intensive AI infrastructure, which is going wherever the grid can support it. For London’s tech economy, this means a slower shift toward a more distributed model. The training will increasingly happen elsewhere. The applications, the interfaces, and the businesses that depend on them will continue to be built in the capital.
For the data centre sector, the implication is straightforward. The question is no longer where the customer wants to be. It is where the power actually is.
