BNEF Finds Solar and Batteries Cheapest US Power Option Regardless of AI Demand Trajectory
BloombergNEF analysis shows solar-plus-storage economics hold even if data center growth slows, but grid bottlenecks and China supply chains cap near-term deployment.
Solar and battery storage remain the cheapest and fastest way to add power capacity in the United States even if artificial intelligence data center demand falls short of the most bullish projections, according to BloombergNEF analysis published on May 19 (2026-05-19). The finding reframes the debate around AI-driven power demand: the technology's economics, not the scale of demand growth, are now the dominant variable for new US capacity additions.2
For energy investors, that severs the link between AI demand forecasts and the investment case for renewables. If solar wins on cost regardless of whether data centers expand at peak projections, developers and utilities face the same build-out incentives under a range of demand scenarios.1
BNEF projects US installed solar capacity reaching roughly 737.8 GW by 2035, up from approximately 231.4 GW in 2024, with total renewable capacity exceeding 1 terawatt — more than double the approximately 414.5 GW installed as of 2024. The IEA estimated in a global energy assessment published around the same period that data centers already account for about half of US incremental electricity demand growth, a share that gives the sector outsized influence over which generation technologies get built.1,6
Yet cost leadership in new capacity does not translate straightforwardly into dominance of data center power supply. BNEF expects fossil fuels — primarily gas and coal — to provide 51% of incremental electricity generation for data centers through 2050, given their ability to generate around the clock regardless of weather. Solar and batteries address intermittency in some configurations, but the economics worsen materially as storage duration requirements increase.2
Battery storage firms are confronting two structural constraints that limit how quickly they can capitalise on demand. Grid connection queues in the US stretch years, and the supply chain for battery packs remains heavily concentrated in China, creating procurement risk and price exposure that utilities and developers cannot easily hedge. High battery pack prices and global shipping bottlenecks were cited by panelists at the BloombergNEF Summit in New York in April (2026-04-xx) as the primary near-term constraints on deployment.4,5
Google's decision to incorporate $1 billion worth of 100-hour batteries from Form Energy into a recent data center project illustrates the direction of travel for long-duration storage, though the economics of iron-air and other long-duration chemistries remain far from proven at utility scale. One project does not establish a market.2
The grid bottleneck cuts across both renewables and data centres. US grid engineers and utility executives described in reporting published on May 19 (2026-05-19) a permitting and interconnection queue system unable to absorb the speed of data center build-out, with estimates pointing to roughly 5,000 miles of additional high-voltage transmission required to support incremental load growth. That constraint affects gas and renewables alike — no new generation technology scales easily if transmission cannot carry it to load centres.6
The Economist reported in May (2026-05-19) that rising US electricity bills are being attributed publicly to AI power demand, though the actual contribution of data centres to retail rate increases is modest against the backdrop of broader grid investment requirements. The political attribution may shape regulatory and permitting decisions even if the economic attribution is overstated.3
What remains unresolved is whether the US permitting system can accelerate fast enough to allow the cheapest capacity — solar-plus-storage — to be built at the pace BNEF's 2035 projections require. At current interconnection queue clearance rates, the gap between least-cost economics on paper and deployed capacity on the ground is wide. The degree to which that gap narrows over the next three years will determine whether the AI power build-out is predominantly gas-fired or renewables-led, regardless of what BloombergNEF's cost curves say.6,5