BNEF's 42 GW forecast gap puts US data centre power buildout in question
BloombergNEF's two 2030 demand scenarios diverge by more than the capacity of several regional grids, complicating investment decisions across US power markets.
Two BloombergNEF scenarios for US data centre electricity demand in 2030 differ by 42 GW, the firm's analysts said on Wednesday (2026-07-22) — a spread wide enough to change what gets built, what gets connected, and what gets paid for across every US power market before the end of the decade.5
BNEF puts total US data centre demand at 207 GW by 2033. Senior associate Nathalie Limandibhratha said the firm has added roughly 100 GW of project capacity to its tracking in the past year alone, and almost every US region ended 2025 with more data centre capacity than BNEF had anticipated. Texas showed the largest gap between forecast and actual buildout.5
BNEF's base case sits in the middle of third-party estimates, which the firm describes as conservative. Bank of America analysts put additional US electric load at roughly 125 GW over the period. The gap between those two figures implies different interconnection timetables and a different set of winners among generation developers.5
Canary Media's analysis of the buildout found BNEF's earlier December work suggested nearly double the load the firm now expects the grid to accommodate.4 If the load growth is real but connection capacity is not, the constraint on how fast AI capacity comes online shifts from technology to infrastructure.
The Electric Power Research Institute has a separate estimate: data centres could account for 9% to 17% of US electricity supply by 2030, or up to 790 TWh, against roughly 4% in May 2026. The wide band reflects genuine disagreement about server efficiency, utilisation rates, and how much of the announced pipeline ever completes.1
The US already leads all other countries in data centre power consumption, China included. In 2025, nearly 40% of global data centre power demand was concentrated in the US, according to Canary Media.3 That concentration is why the forecast debate is primarily a US story, even as the supply chain for batteries, fuel cells, and transformers is global.
On-site generation is where the grid shortfall gets priced. Rystad Energy projects fuel cell market revenues rising from around $2.8bn in 2025 to roughly $30bn by 2030, driven by developers turning away from congested grid queues toward on-site power.2 Rystad's projection is not a booked order book, and the fuel cell industry has missed projections before. But the direction is unambiguous: developers are paying a premium for speed over grid connection timelines.
Batteries show the same dynamic. The US added a record 57.6 GWh of new storage capacity in 2025, according to the Solar Energy Industries Association, bringing total deployed capacity to 166.1 GWh. The group projects annual deployments reaching 110 GWh by 2030, with a significant share driven by data centre demand.1 Fluence says it is engaged in more than 30 GWh of data centre-related projects globally, with a meaningful US portion, according to CEO Julian Nebreda. Tesla booked $430m in revenue from storage sales to xAI last year, and Calibrant Energy has agreed to supply a 31 MW/62 MWh battery system at an Aligned data centre campus in the Pacific Northwest.1
Grid queues remain the primary bottleneck. Battery storage firms see surging interest from AI data centres, but lengthy interconnection queues and a supply chain heavily dependent on China are limiting how fast the industry can scale, Reuters reported on Monday (2026-05-18). The queue is not primarily a technology problem. It is permitting, transformers, and skilled labour — each of which moves on years-long timelines.1
PJM has already priced the load in its most recent capacity auction. Grid operator PJM announced in the week of 2026-07-13 the results of a power procurement auction expected to add billions in electricity costs across 13 states as data centre demand rises. Capacity prices are the clearest market signal that the demand is real, even where the long-run forecast is disputed.6
The next material signal is whether BNEF's 42 GW scenario gap narrows in its coming forecast revisions — and which direction it moves. If the conservative base case holds, fuel cell and battery projections may already be factored into equipment order books. If the upside scenario proves correct, the constraint shifts to transformers, interconnection slots, and gas turbines, and projects with secured grid capacity become the scarce asset.5