The electricity demand driven by AI data centers is expanding more rapidly than anticipated, shifting the core supply challenge from "is there enough power" to "can it be built in time".

Fresh analysis from Goldman Sachs warns that AI's incremental power consumption from early 2024 through the end of 2030 will equal the entire electricity usage of Japan, the world's fifth-largest power consumer. Meanwhile, shortages of turbines, transformers, transmission lines, and skilled labor, compounded by local political resistance, have become the real obstacles to bringing supply online.

The bank has raised its US power demand growth forecast from a compound annual rate of 3.2% to 3.5%, and lifted its 2030 projection for US data center electricity consumption from 83 gigawatts to 108 gigawatts. Globally, the expected increase in data center power demand versus 2025 levels has been revised sharply upward from 117% to 170%. In tandem, capital expenditure forecasts for hyperscale cloud providers have been upgraded, with 2027 estimates rising from $1.2 trillion to $1.7 trillion, and 2029 projections climbing from $1.5 trillion to $2.1 trillion.

Goldman Sachs emphasizes that its base case does not anticipate widespread blackouts across the United States, but regional power strain, rising electricity rates, and public opposition will profoundly influence actual construction timelines. The PJM interconnection territory is already operating at critical tightness, while ERCOT is expected to come under pressure around 2028 as load levels climb. Whether projects secure public and political backing will largely determine if the forecast capacity can actually be built.

Demand Keeps Rising: Efficiency Gains Overshadowed by Broader Adoption

Goldman analyst Brian Singer notes that improvements in chip and model efficiency are genuine but have not led to an overall contraction in computing expenditure. Cheaper compute unit prices have expanded the range of economically viable applications, with greater token generation volumes, wider agent-based workloads, and persistently growing hyperscaler budgets collectively offsetting and exceeding the efficiency dividend.

Vacancy rates across major US data center markets have collapsed from 2% to 7% down to just 1% to 2%, with Goldman expecting a recovery to roughly 3% by 2030. This tightening in supply-demand dynamics further bolsters the case for upgrading demand forecasts.

Singer quantified the trend with a striking frame of reference: "If you start counting from early 2024 to the end of the decade, over that seven-year period, the additional electricity consumed by AI equals the entire country of Japan—the world's fifth-largest power consumer. The scale is enormously significant."

Supply Pathway: Gas Leads, Renewables Follow, Nuclear Arrives Later

Goldman outlines a phased roadmap for data center power supply: near-term reliance on simple-cycle gas turbines alongside renewable energy paired with storage, a mid-term transition to combined-cycle gas units, and long-term expectations placed on nuclear power, with the restart of select existing nuclear plants serving as a bridge. By 2030, the data center power mix is projected to comprise roughly 60% natural gas and 40% renewables including storage.

With grid interconnection queues stretching two to seven years, behind-the-meter (BTM) gas generation has emerged as a critical workaround. Goldman has raised its BTM gas capacity forecast to approximately 30 gigawatts by 2030, translating to over 20 gigawatts of actual output, representing about 20% of total data center demand at that time. The bank positions BTM as a transitional solution rather than an endpoint, noting that large customers will ultimately prefer grid connection for lower costs and higher reliability.

Regionally, the rise of MISO (Midcontinent Independent System Operator) carries structural significance—regulated utilities there can offer a one-stop package spanning generation, transmission, distribution, and regulatory relationships, with electricity rates determined by state-level regulators rather than wholesale markets, helping to manage the political sensitivity of rate increases.

Four 'T's and Seven 'P's: Bottlenecks Shift From Megawatts to Supply Chains and Politics

Goldman analyst Allison Nathan distills the core constraints into four 'T's: turbines, transformers, transmission, and tradespeople. One major turbine manufacturer has indicated it expects to sell out half of its 2031 capacity by year-end, with equipment lead times profoundly shaping choices about power sources. Meanwhile, after a decade of stagnant US power load growth, shortages of electricians and high-voltage welders have become increasingly acute, with long training cycles creating structural imbalances that cannot be resolved quickly.

Singer further proposes a seven 'P' framework covering AI pervasiveness, productivity of chips and models, price of power, policy, parts, people, and the physical environment. Among these, physical environment risks deserve particular attention: over half of data center sites are located in high-physical-risk areas facing heat, humidity, and drought challenges. The water-versus-power tradeoff from cooling systems is most acute in West Texas, where securing water access is harder and costlier than adding new electricity.

Political Resistance May Be the Ultimate Constraint: Over 300 Local Bans Already in Place

Goldman analyst George Lee characterizes interconnection as "the single most important issue" facing the US utility sector, noting that more than 300 local and regional moratoriums are already in effect, while state-level bans remain comparatively limited. He warns that if the AI and power industries fail to develop a coordinated public communications strategy, political factors will become the ultimate hard constraint.

Community opposition centers on five primary concerns: blackout risk, rising electricity bills, water consumption, noise pollution, and waste heat emissions. Goldman's working assumption is that projects will migrate toward regulatory-friendly jurisdictions, which could further intensify geographic concentration. At the same time, competition among towns vying for tax bases, construction jobs, and infrastructure investment is providing landing space for some projects.

On utility balance sheets, capital expenditures across Goldman's covered regulated utilities are expected to grow approximately 60% over the next five years compared to the prior five-year period. Incremental financing is projected to come 30% to 50% from equity issuance, with leverage ratios retaining roughly a 100-basis-point buffer below downgrade thresholds.