BNEF's 42 GW Scenario Gap Leaves US Data-Center Demand Forecast Unresolved
BloombergNEF's own 2030 demand models span a 42 GW range as Bank of America estimates data centers could add 125 GW of US electric load.
Two of BloombergNEF's scenarios for US data-center electricity demand by 2030 diverge by 42 GW, a spread wider than the installed generating capacity of most regional US grids. BloombergNEF senior associate Nathalie Limandibhratha flagged the gap at a presentation on Wednesday (2026-07-22), noting the firm has tracked roughly 100 GW of project capacity added in the United States over the past year. Texas recorded the largest divergence between forecast and actual buildout of any state.4
Bank of America analysts put the potential US electric load addition from data centers at roughly 125 GW by 2030, while JLL projects developers may require roughly 100 GW of new data-center capacity over the same period. BNEF describes its own base case as relatively conservative, sitting in the middle of third-party estimates — the upside scenario implies a demand surge that would test the interconnection queue in most regions.4,3
Almost all US regions ended 2025 with more data-center capacity than BNEF had anticipated. Virginia shows what concentrated demand looks like: commercial electricity sales there rose by nearly 30 million megawatt-hours between 2019 and 2025, with EIA attributing much of that growth to the state's data-center cluster. The US now accounts for nearly 40% of world data-center electricity, according to Forbes.5,64
McKinsey estimates AI-related infrastructure spending could exceed $5 trillion by 2030, encompassing not just chips and buildings but the turbines, transformers, and transmission lines needed to run them. Some developers are already routing around the utility queue. Crusoe signed a $1.25 billion contract with Boom Supersonic for 29 jet-engine turbines to power data centers across the country, while startup Panthalassa raised $140 million to build buoy-like floating devices that generate power as they move with the waves.3,2
NYMEX Henry Hub front-month gas closed at $2.83/MMBtu on 2026-09-12, and the contrarian read on that price is bearish, driven by demand. Gas is not pricing in a demand shock.4
Part of the explanation is arithmetic. Even a 4-6 GW data-center complex consumes only about 1 billion cubic feet per day of natural gas depending on turbine efficiency. Meaningful, but well within the 2-3 Bcf/d seasonal swings the US market routinely absorbs. The gas upside from data centers is real but smaller than the gigawatt headlines suggest.1
Timing adds another constraint. Data centers want power faster than utilities can deliver, pushing some developers toward demand flexibility. Several are aiming to reposition data centers as flexible grid assets that can reduce inference operations during peak demand periods, and some AI models are already moving toward surge pricing. If flexible demand management scales, it reduces the need for dedicated peaking capacity and softens incremental gas requirements.1
The 42 GW spread between BNEF's own scenarios is the best current measure of how much is still unknown: chip efficiency gains, interconnection timelines, developer follow-through. Utility integrated resource plans and interconnection queue data over the next year will clarify how much of the 100 GW of tracked project capacity converts into signed power purchase agreements. If it does not, the AI trade in power markets may be running ahead of the physical buildout.4