BofA Sees 100 GW US Power Gap as AI Build Outpaces Utility Supply Plans
Regulated utilities will add just 93 GW against a 230 GW five-year need, Bank of America says, pushing data center developers toward behind-the-meter gas generation.
Bank of America analysts published a sharp capacity arithmetic on Thursday (2026-07-17): the United States will need more than 230 GW of new generating capacity over the next five years to absorb AI-driven electricity demand, but regulated utilities are expected to contribute only around 93 GW of accredited supply, leaving a gap exceeding 100 GW.5
The gap feeds directly into gas demand. The fastest path from shortfall to first electrons is on-site combustion turbines fired on NYMEX Henry Hub front-month gas, at $2.91/MMBtu as of the week's close. More than 7.5 GW of data center projects with on-site generation are already under construction, BofA's Global Research report said, with a further 60 GW-plus in pre-construction — a pipeline largely invisible to grid operators working from utility interconnection filings.5
BofA analysts flagged that planned generation additions likely overstate available supply. Intermittent wind and solar contribute substantially less accredited capacity during peak demand periods than their nameplate ratings imply, and hyperscalers seeking firm power commitments cannot take nameplate at face value. That mismatch sits at the centre of why utility counterparties are struggling to sign the 24/7 reliability agreements data center operators want.5
As a result, BofA expects more data center developers to move toward behind-the-meter generation at pace. Engine manufacturers Caterpillar, INNIO, Rolls-Royce and Wärtsilä have already expanded production to meet rising demand for distributed gas turbines, the analysts noted. The shift is commercially rational: demand from hyperscalers is sufficiently inelastic that a 10% increase in real electricity prices typically produces only a 1% to 2% decline in consumption, according to BofA's research, meaning large users absorb cost increases rather than curtail load.5
The analysts summarised the grid predicament plainly: "The market is no longer constrained by demand — it is constrained by where power can actually be delivered."5
Stock markets had already begun registering this reading. Fluence Energy shares closed at $24.16 on Friday (2026-05-08), up 98.2% in a single week, after the company disclosed master supply agreements with two hyperscalers and a record $5.6 billion backlog, according to Quick Read Capital. The move illustrated how quickly capital rotates toward infrastructure that demonstrably connects to AI power supply chains.1
The M&A market has registered the same shift more broadly. Reliability has become the primary driver in energy, utilities, and resources deal flow, with investors redirecting capital from upstream exploration toward grid and gas assets that can deliver contracted, uninterruptible power, according to Asian Power analysis from June (2026-06-23).3
Dr. Anas Alhajji, speaking on Macro Voices, named natural gas and LNG as the primary market beneficiaries of AI power demand, with coal gaining secondary interest as a cheap domestic fuel in markets where gas supply is stretched. That view fits the BofA supply gap scenario: behind-the-meter gas turbines remain the fastest technology to deploy from pre-construction to live generation.4
The IEA's World Energy Outlook 2025, published in May (2026-05-20), treated AI-driven electricity demand as a fresh energy security risk, calling for greater supply diversification rather than concentration in any single technology. The BofA 100 GW number gives that warning a concrete order of magnitude.2
NYMEX Henry Hub front-month at $2.91/MMBtu sits well below levels that would deter distributed gas generation investment. The real test is whether the 60 GW-plus pre-construction data center pipeline converts to operating capacity before the accredited supply gap widens further, and whether the nameplate-to-firm discount on wind and solar proves larger than current interconnection studies assume.5