人工智能的崛起：能源与经济影响的实证审视

Mark P. MillsExecutive Director, National Center for Energy Analytics Distinguished Senior Fellow, Texas Public Policy Foundation A RESEARCH PARTNERSHIP BETWEEN:The National Center for Energy Analyticshttps://energyanalytics.org The Hamm Institute for American Energyhttps://hamminstitute.org Executive Summary Total private sector spending committed to artificial intelligence (AI) isnow a significant share of U.S. GDP. Spending on building data centers alone—exceeding $50 billion annually and rising—has now surpassed spending on all less than about 75 GW, possibly as much as 100 GW of generation by 2030. Andneither outcome includes the electricity demands for expanding the ancillarybut directly related telecommunications networks, as well as that needed for The evidence of an emerging structural shift in the U.S. economy can beseen in the combination of the epic spending on AI, rapid adoption of AI tools, To meet that much new demand by 2030, underlying engineering realitiesshow that most of the additional electricity generation will necessarily come The core consequence of AI deployed at scale is in its potential for boostingproductivity, the feature of every economy that drives growth and prosperity. Ifdemocratizing AI elevates U.S. productivity growth only to the average of thepast half-century, it will add a cumulative $10 trillion more to the U.S. GDP than That rise in gas demand will occur contemporaneously with roughly thesame amount of new demand coming from additional LNG export terminals The nation is capable, technically, of meeting such a level of growth innatural gas production, pipeline installations, and power plant construction. Often ignored in AI forecasts: the additional wealth created by using thenew technology leads to behaviors that use more energy. An extra $10 trillionwould lead to increased overall energy use equivalent to about five billionbarrels more energy over the next decade. Such a wealth-induced increase in In the longer-term, as the AI structural revolution continues to play outpast 2030, even greater energy demands will emerge to power the next phaseof growth. Those demands will likely be met increasingly from nuclear energyand additional solar capacity. But each of those require associated infrastruc-ture expansions that, inherently, take far longer than the current torrid pace AI data centers are not unique in that regard. Energy is the “master resource”needed for operating every part of civilization. As Nvidia CEO Jensen Huangrecently said: “AI is energy, AI is chips, the models, and the applications . . .. And we need more energy.” As it happens, data centers are measured and The challenge is that electricity-related energy policies now in place wereframed during the recent period of low or no growth, combined with misguidedpursuits of an energy transition to replace conventional energy sources. TheAI boom has illuminated the fact that new periods of growth are inevitable— What is unique about AI data centers is the scale and velocity of powerdemands now emerging. Some individual data centers now under constructionwill have city-scale power demands, and hundreds are being built or planned.The rate of construction—especially in combination with reshoring manufac- Google observed earlier this year that “AI presents the United States with agenerational opportunity for extraordinary innovation and growth” but that itrequires “expedited effort to increase the capacity of . . . U.S. energy systems.”A July 2025 White House directive,Winning the Race: America’s AI Action Plan, Policymakers, investors, and businesses in the AI supply chain are allinterested in discerning just how much additional electricity will be neededspecifically for powering data centers and how it will be supplied. For this The facts and trends suggest AI digital demands will require building no Key Takeaways with Implications for Policymakers A key, enabling requirement: meet the unprecedented scale ofelectricity demand of individual facilities, at the velocity being built. The great AI inflection is market-driven:•The race to build AI infrastructures and services is being funded by the private sector, i.e., it is not a policy-driven transformation.•The pace of AI adoption depends on integration challenges for specificapplications, and the maturity and efficacy (power) of the AI tools, i.e., •Hundreds of planned data centers have demand exceeding 300 MWeach, many over 1 GW, with construction completions often in two orthree years. There are no precedents in utility history for such scale orvelocity.•The pace and scale are in tension with supply chains and workforce Three key metrics illuminate the transformation:•The rate of progress in the underlying AI compute capabilities follows a well-established, predictable trend of exponential gains.•AI computational performance has increased athousandfoldsince Engineeringrealities will dictate viable solutions.There are twotimeframes fo