The UK government's marquee AI infrastructure project — a "sovereign" supercomputer in Loughton, Essex, built by startup Nscale — will miss its 2027 launch date by years. UK Power Networks has told the company the local grid cannot supply sufficient electricity until the early to mid-2030s. The site, which was still an operational scaffolding yard when the Guardian first investigated in March, now faces the more fundamental problem of physics: there is not enough power to turn on. The numbers frame the absurdity. The Loughton facility needs up to 90MW — equivalent to the consumption of roughly 315,000 homes. That is one project. Ofgem revealed in June that 315 datacentres are queued for National Grid connections, representing 73GW of aggregate demand. Britain's entire peak electricity demand is 45GW. The queue alone exceeds the country's capacity by 62%. Nscale, which is seeking a $35bn (£26bn) flotation this year, says it is "committed to the project" and is exploring on-site power generation and ways to accelerate the connection process. These are not small pivots. On-site generation at 90MW scale likely means gas turbines or dedicated renewables with storage — each carrying its own planning, permitting, and capital timeline. The grid connection queue is a structural problem, not a negotiation. The political exposure is significant. Keir Starmer's government cited Loughton specifically when announcing its AI strategy in 2025, branding it the "largest UK sovereign AI datacentre." That language — sovereign — implies national capability, strategic independence from US hyperscalers. A decade-long delay turns a sovereignty play into a sovereignty gap, precisely during the window when compute capacity is being locked in globally. UKPN's statement is telling: many datacentres, while connecting to the distribution network, are "awaiting works on the wider national electricity transmission network." Translation: even if the local wires are ready, the national backbone cannot deliver. This is not a Loughton problem. It is a systemic bottleneck that applies to every large energy consumer trying to connect. Nscale's $35bn flotation ambition sits uncomfortably alongside the reality. The company is asking public markets to value it at hyperscaler scale while its flagship site has no confirmed power date and its primary asset was a scaffolding yard six months ago. The gap between valuation narrative and operational readiness is wide enough to drive a datacentre through. The deeper pattern is one Britain has repeated across infrastructure for decades: announce ambitious targets, underinvest in enabling systems (grid, planning, transmission), then express surprise when timelines slip. AI compute is energy. Energy requires grid. Grid requires investment made years before demand arrives. The UK made none of those investments and is now discovering that sovereignty cannot be declared into existence.
