Grid bottlenecks and supply chains threaten to delay the AI power surge
Wood Mackenzie's 8% U.S. substation deficit and a China-dependent battery supply chain complicate near-term forecasts for AI data center electricity demand.
NYMEX Henry Hub front-month fell to $2.97/MMBtu on Wednesday (2026-09-23), down 2.62% on the session, even as Big Tech continued to accelerate spending on AI data centers that will ultimately require substantial gas-fired backup generation. The divergence between gas market pricing and the AI demand narrative has widened.7
The Electric Power Research Institute projects data center electricity consumption could reach 9% to 17% of total U.S. supply by 2030, or up to 790 terawatt-hours annually, compared with roughly 4% as of mid-2026. Goldman Sachs and BloombergNEF have both revised their data center power forecasts higher. Virginia alone saw commercial electricity sales rise by nearly 30 million MWh between 2019 and 2025, with EIA attributing much of that to the state's concentration of data centers.1,6,5
The demand trajectory appears firm. Grid readiness is harder to model.7
Wood Mackenzie estimated in early September (2026-09-03) that the U.S. faces an 8% substation deficit. That shortfall means roughly one in twelve required grid connection points is unavailable to support new large load additions. Substations take years to permit, procure, and install, and the manufacturing backlog for grid transformers has been widely reported.7
DataM Intelligence noted in July (2026-07-16) that power availability and equipment lead times are already reshaping AI infrastructure deployment schedules. Long-term capacity contracts are being signed, capital is available — but physical connection dates are slipping. Ambition and capital are abundant. Physical grid infrastructure decides when a data center actually draws power.3
Battery storage has emerged as the favored workaround: co-locate behind the meter, absorb solar generation during the day, discharge at peak, and sidestep interconnection queues. The U.S. added a record 57.6 GWh of new battery energy storage in 2025, according to the Solar Energy Industries Association, bringing total deployed capacity to 166.1 GWh. Fluence CEO Julian Nebreda said his company is engaged in more than 30 GWh of data center-related projects globally. Tesla generated $430 million in revenue last year selling storage systems to Elon Musk's xAI.1
Yet the supply chain underpinning that buildout is heavily exposed to China. Battery storage firms told Reuters in May (2026-05-19) that lengthy grid connection queues and Chinese supply chain dependency are both constraining their ability to scale. The SEIA projects annual battery deployments reaching 110 GWh by 2030, assuming no meaningful supply disruption. That assumption grows harder to sustain.1
A further complication comes from Australia. Nextdc's Shayne Kumar argued at a Melbourne energy forum during the week of June 8 (2026-06-08) that AI data centers help reduce power bills by running stable, flat loads that absorb excess solar and lift network utilization. If large-scale flat load demand suppresses duck-curve volatility and compresses peak spreads, the power price upside implied by headline consumption forecasts would be smaller than those numbers suggest. That argument has not been tested at anything close to the scale now being discussed.2
The U.S. already consumes nearly 40% of the world's data center electricity, according to Forbes in August (2026-08-23). That share will grow. But the bull case for power prices and infrastructure investment rests on grid expansion keeping pace with load growth through 2027-2030. Wood Mackenzie's substation deficit and the China-exposed battery supply chain both push against that timing.4,7
JKM spot for Asian LNG stood at $26.05/MMBtu on Wednesday (2026-09-23). Neither Asian gas markets nor NYMEX gas is pricing a near-term surge in electricity demand from the AI buildout. Watch how the Wood Mackenzie substation deficit figure moves over the next two quarters. If interconnection queue clearance rates outpace new applications, the demand timeline firms. If the deficit widens, power market investors positioned for 2028 load growth may find themselves considerably early.7,1