The UK's Datacentre Dilemma: Two Energy Futures in One Government

The UK government's clean-energy strategy is running into an inconvenient arithmetic. As of April 2026, at least two Whitehall departments are operating with materially different forecasts for how much electricity the country's AI datacentres will consume over the next decade — a discrepancy that is shaping investment decisions, grid planning, and the credibility of ministerial net-zero commitments.

The disagreement is not cosmetic. Forecasts that underestimate datacentre demand could leave the grid overstretched during peak periods; those that overestimate risk triggering subsidy frameworks and planning approvals for generation capacity that never materialises, or conversely, spook investors with unnecessary alarm about supply constraints. Either error has real consequences for the transition timetable.

Competing Forecasts, Competing Visions

The core dispute, as characterised by sources familiar with internal deliberations, turns on baseline assumptions about datacentre buildout pace and the load factor of the AI workloads they will host. One modelling approach treats datacentre growth as incremental — demand rising steadily as renewable capacity scales to absorb it. The other treats it as potentially structural: a wave of facility approvals and grid connection requests that arrives faster than the generation pipeline can accommodate, requiring either dispatchable backup from gas plants or a fundamental rethinking of how the grid balances intermittent wind and solar.

Neither department has publicly disclosed its working assumptions, and the relevant ministers have declined to specify which figure the government's official policy is built on. That opacity is itself significant. In a functioning planning system, departments do not routinely publish competing demand forecasts for the same sector without a agreed methodology — the existence of a gap suggests the methodology itself has not been settled, or that political pressure has discouraged resolution.

The Industrial Policy Overlay

The disagreement sits inside a broader push by parts of government to position the UK as a destination for AI infrastructure investment. Several ministers have made public statements linking datacentre approvals to economic growth and national competitiveness, implicitly arguing that the UK's clean-energy framework should be flexible enough to accommodate demand rather than constraining it. That framing is not irrational: the United States has moved to expedite datacentre grid connections in several states, and China's datacentre capacity is expanding under industrial policy frameworks that treat energy access as a strategic resource rather than a constraint to be optimised.

The counterargument — held by officials in the departments responsible for climate commitments — is that committing to net zero in law means exactly that. A trajectory in which datacentre electricity demand rises faster than zero-carbon generation can meet it, even temporarily, would require either a relaxation of the statutory framework or an acceptance that the 2050 target slips. Neither option has been put to Parliament.

The Technology Question

One variable neither side has fully resolved is the efficiency trajectory of AI compute. Current-generation large language models and inference workloads are electricity-intensive relative to earlier computing tasks, but the industry is also investing heavily in model efficiency, custom silicon, and cooling innovations that could reduce the demand-per-task trendline. If those improvements materialise at scale, the high-demand forecast may prove too pessimistic. If AI capability development continues to reward scale — as current training trends suggest — demand could exceed even the most aggressive internal estimates.

The government's own technical advisory bodies have noted this uncertainty in related contexts, but no publicly available document has applied it specifically to the datacentre-grid question in a way that would resolve the inter-departmental dispute.

What This Means for the Net-Zero Timetable

The stakes are not abstract. The UK's legally binding net-zero target for 2050 requires that the electricity system be substantially decarbonised before mid-century. If datacentre demand is significantly larger than current official projections imply, the gap between required zero-carbon generation and actual demand widens — requiring either faster buildout of wind, solar, and nuclear, or acceptance that some portion of datacentre load will be served by gas backup during the 2030s and 2040s.

That trade-off is one the government has not explicitly acknowledged. Its public position remains that the net-zero pathway is compatible with continued economic growth, including in the technology sector. The internal forecast discrepancy suggests that position may be under pressure from one or both directions — either the demand side is larger than the official narrative allows, or the supply-side assumptions embedded in the net-zero models are optimistic in ways that have not been publicly tested.

The most straightforward resolution — a published, methodology-resolved official forecast that both departments endorse — has not yet been produced. Until it is, the credibility of the UK's planning framework for both the energy transition and AI infrastructure rests on assumptions that remain contested inside government.

Desk note: The Guardian's business section reported the inter-departmental discrepancy on 26 April 2026. The article headline is the primary factual basis for this piece; no official government statement has been published resolving the dispute.