The Federal Reserve Bank of Dallas convened the Powering AI conference on March 4–5, 2026, to examine whether U.S. electricity infrastructure can support the rapid buildout of AI data centers. Conference participants concluded that the technical capacity, capital and growing demand flexibility exist to power AI in the near term, but coordination problems—timing mismatches, supply constraints, labor shortages, financing uncertainty and community opposition—pose the biggest threats to meeting growth at the required speed and scale.
Construction timing and near-term capacity
Panelists highlighted that the most immediate obstacle is timing rather than sheer volume. Data center construction can take about two years, while new power plants commonly require twice that time and transmission projects can take 7–10 years, creating planning mismatches. Still, conference participants said existing generating capacity can help meet near-term demand through higher utilization: Texas fossil fuel plants, for example, operate at less than 50 percent capacity over the course of a year but can scale to 70 percent or higher utilization. That means some headlines suggesting imminent shortages may overstate near-term risk, even though adding large AI loads could push up variable and generation costs and affect prices.
Flexible data centers and reliability
Speakers noted that rising peak demand is a central reliability concern but that data centers can become part of the solution if they adopt flexible operating models. Traditional cloud data centers historically requested roughly 30 MW, took years to reach full capacity and ran at relatively low utilization. New AI facilities may differ and could evolve from inflexible baseload users to flexible loads that respond to price signals and grid conditions. The January 2026 Winter Storm Fern illustrated how commercial and industrial curtailments helped prevent price spikes and kept demand below forecasts during peak stress. Some regions are also requiring or favoring bring-your-own-power arrangements, on-site generation, curtailment commitments and workload shifting to speed interconnection.
Supply constraints beyond generation
Conference participants emphasized that supply constraints are multifaceted: generating capacity, long backlogs for electrical components, rising construction costs and financing frictions all play roles. Demand uncertainty compounds those constraints, because utilities and investors need firm commitments to separate viable projects from speculative ones. Panelists described interconnection queues and speculative land deals as key risks to efficient infrastructure allocation.
Skilled labor shortages
A widespread shortage of qualified trades and technicians surfaced as perhaps the most serious constraint. Panelists warned of insufficient electricians, technicians, network engineers and specialized workers across solar, battery, data center and gas generation projects. Some companies are investing directly in reskilling, and speakers suggested that higher wages could attract more workers if roles are better promoted, but the shortage remains a tangible barrier with no single ready-made solution.
Financing, speculation and the ERCOT queue
Investment flows are large: hyperscalers plan $700 billion to $900 billion in annual capital spending in 2026–2027, with Amazon alone planning a $220 billion capital budget. Panelists generally saw abundant capital as manageable—hyperscalers reportedly generate operating cash flows covering around 80 percent of capital expenditures—but they warned of a $1.5 trillion financing gap that will need diversified debt markets including corporate bonds, private credit and securitization. Investors have moved rapidly from reluctance to active debt discussions, but panelists stressed that utilities and financiers require firmer commitments and stricter criteria to avoid funding speculation. In Texas, ERCOT’s interconnection queue sat at 232 GW in future demand through 2030, while the all-time record demand in ERCOT was set at 91 GW in July 2026. After the conference, ERCOT adopted new rules requiring large loads seeking interconnection to post significantly higher, non-refundable fees, and Texas Governor Greg Abbott outlined related state-level requirements.
Community and political dynamics
Speakers said social and political factors are as consequential as technical ones. Community opposition has blocked billions of dollars in data center investment, driven by fears about higher electricity costs, noise, water use, pollution, grid impacts and a perceived lack of transparency. Panelists noted that public anxiety about AI can make projects fragile: while some see AI as transformational, others point out that only about 5 percent of companies piloting AI achieve scaled value. Geographic shifts in siting—from city centers to unincorporated areas—may ease urban resistance but could transfer impacts to rural communities with fewer local zoning controls. Multiple panelists urged stronger community engagement and clearer conversations about actual versus perceived employment impacts.
Powering AI: a coordination challenge
Across sessions, the recurring theme was coordination. Conference participants agreed the nation has much of the technical capacity, capital and growing demand flexibility to support AI growth, but unlocking that potential requires agility, transparent commitments, regulatory reform, workforce development and community engagement. The critical question for policymakers, utilities, investors and industry is whether stakeholders can align timelines, financing and social acceptance rapidly enough to deploy infrastructure at the scale AI buildout demands.
