America is hurtling towards an energy transformation. The U.S. government and private enterprises are investing $500 billion over several years to build new artificial intelligence manufacturing hubs, data centers, and compute clusters. Several of these will demand more power than the New York City metropolitan area.¹

The race to build energy infrastructure, a core component of the “AI stack” as critical as chips or data, is inexorably advancing. Training and running frontier AI models now consumes electricity comparable to hundreds of thousands or even millions of U.S. homes.²

Still, energy buildout today faces many challenges: lengthy timelines, atrophied innovation, avarice and mismanagement by utilities companies, as well as structural incentive misalignments across the grid.³

Policymakers and hyperscalers — the companies advancing AI through colossal compute investment—must consider how to build this infrastructure such that both they, and the Americans they are building next to, benefit.

Some hyperscalers are beginning to take steps, though their success metrics still center largely on construction pace and cost. But achieving these goals necessitates also addressing public perception, historical precedent, and federal policy.

The average American is less supportive of data center construction projects compared to energy buildouts such as natural gas or nuclear. Grievances include excessive water consumption and electricity bills, and in some cases, have mobilized local opposition.⁴ In August 2025, Tuscon residents rejected Amazon’s behemoth Project Blue data center campus during an ongoing drought.⁵ Similar resistance movements have cropped up across the country, often with grassroots origins. To build public trust, hyperscalers must demonstrate their myriad mutually aligned interests.

There are numerous realizable benefits. To start, building more local generation could reduce local consumers’ energy costs. Simultaneously, this unprecedented injection of billions of dollars of private investment can modernize the grid, its power supplies, and connecting lines—along with supporting new R&D. Finally, strong and stable energy infrastructure is key to continued U.S. leadership in AI and a fundamental pillar of American national security.

Implemented poorly, these initiatives will silo benefits from consumers, further entrench existing energy inefficiencies, and exacerbate hyperscalers’ trust deficit. Executed well, the ongoing buildout could not only reverse America’s AI-related energy challenges but pose an immense opportunity for both public and private energy prosperity.

But to be done well, construction must be done.

The Transmission Challenge and Interconnection Opportunity

Improved transmission — the network of power lines, substations, and transformers transporting energy over long distances — is commonly cited as a remedy for America’s swelling energy requirements. Yet despite that consensus, building transmission is challenging.

Between 2023 and 2025, $64 billion worth of data center initiatives were blocked or delayed due to transmission constraints, buried under legal permitting standoffs and “not-in-my-backyard” (NIMBY) activism, according to Data Center Watch. “The trend is clearly accelerating,” a representative shared. Two recent legal stoppages in Georgia and North Carolina, representing $47 billion cumulatively, independently support that claim.⁶

Attributing the problem to NIMBYs alone is an oversimplification. Inherently, long-range transmission means cheaper energy in the importing region at the cost of raising expenses for consumers in the exporting area. As such, local decision-making regulators often resist these projects to protect their constituents from price hikes.

Caught in the middle are communities bearing the presence of transmission built near them without the economic upside. Their legitimate concerns can create structural resistance to transmission expansion beyond simple neighborhood opposition.

One solution involves looking past state lines. “Federal action can be a real driver for these challenges,” says Ajey Pandey, researcher at Semianalysis. “Most utility regulation happens at the state level, and states can often be territorial about demanding the benefits of infrastructure while insisting another state pays for it. FERC has a lot of leeway to make big changes in the name of ‘just and reasonable’ rates, and they deserve support in picking necessary fights for interstate transmission.”⁷

The trouble with transmission is that without such policy changes, the incentive alignment is not there. This challenge applies specifically to long-distance lines that move power across states. Interconnection is a different part of the system: it deals with shorter links that hook up new generation sources to the grid. They are local, do not require multi-state coordination, and face less formidable political and regulatory obstacles.

Interconnection builds both upside potential and downside protection for hyperscalers. By connecting “behind the meter” projects — localized energy systems — to the grid, during demand-side blips consumers have access to excess energy, and hyperscalers can more easily sell it. This is in addition to new manufacturing jobs, increased self-sufficiency, and reindustrialization synergies.

Interconnection is already a priority outside of the U.S. This summer, the authors visited the construction site for South Korea’s flagship Yongin industrial project, which upon completion, will be a supercluster of co-located chip foundries and generation plants located south of Seoul. Rather than treating “interconnect” as an afterthought, Korea Electric Power Corporation (KEPCO) has baked grid expansion into its national industrial strategy.

^{Construction on the new Yongin complex, which upon anticipated completion in 2030 will host six fabs and three power plants, including a 1.05GW LNG combined heat and power plant. Credit: Seth Goldin.}

By proactively building substations within the semiconductor cluster itself, and connecting them to a $53.5 billion, 15-year transmission modernization plan, AI-related buildout can trickle into consumer benefit.⁸ Yongin’s interconnection could be transformative — without powerful interests like hyperscalers pulling their weight, American consumers will not experience similar benefit.

Amazon Web Services (AWS) has begun investing in dedicated substations and new transmission routes as a way to buffer against grid crowding, while Google, Amazon, and Microsoft have urged the Irish government to let them build “private wires” connecting data centers directly to solar and wind farms.⁹ However, these investments pale in comparison to their spending on power generation and data centers.

Aligning with Americans’ Interests

Beyond grassroots opposition and red tape, interconnection initiatives in the U.S. also often face a more significant structural opponent: utility companies.

Concretely, utilities lack incentives to build transmission that benefits out-of-service-territory customers, which increases local energy costs with marginal upside. DOE partnerships allow data center operators to directly fund these transmission projects through federal channels, bypassing state-level regulators who can singlehandedly block these net-good initiatives.

Left to their own devices, utilities prioritize net-new, long-term construction projects that grow capital expenditure and shareholder returns, often at the expense of ratepayers.¹⁰ This incentive is reinforced by how utilities earn revenue from capacity markets, systems designed to pay generators for guaranteeing electricity will be available during moments of extreme demand—like a heatwave in August when the grid must supply AI data centers, factories, and millions of air conditioners all at once.¹¹

These payments are meant to ensure reliability even when plants sit idle most of the year. “Electric utilities are risk-averse institutions who are reluctant to try technologies like dynamic line rating or demand response, which would improve system capacity on existing infrastructure,” Pandey explained. The result is a structure where utilities profit more from building new assets and locking in long-term payments than increasing grid efficiency.

As a result of how utilities are currently structured, even if a hyperscaler self-finances new generation infrastructure and the associated interconnection, that price could trickle down to consumers.¹² Better interconnection reduces these capacity costs by enabling more competition among regional generators. While utilities are disincentivized to support this, the Federal Energy Regulatory Commission’s (FERC) EL24-80 proceeding is currently reviewing whether utilities should retain unilateral control over financing interconnection upgrades.¹³

These misaligned incentives hide a counterintuitive truth: more demand does not necessitate higher local energy prices. A local system’s “load factor,” a measure of how evenly electricity is consumed over time, is a more meaningful indicator of prices, with higher load factors generally leading to lower costs.

In fact, high load-factor customers like data centers can actually help utilities because the grid’s biggest costs — including power plants and transmission lines — are built to meet peak demand, not average demand. Hyperscaler grid involvement spreads across more usage, which can bring average rates down.¹⁴

Maybe-In-My-Backyard

To truly be “good for consumers,” AI buildout cannot simply reduce local energy costs; new infrastructure also must avoid harming the communities it operates in.

There are legitimate claims of buildouts — such as xAI’s Tennessee Colossus supercluster — polluting the local environment and draining water resources.¹⁵ These realities have emerged from companies’ desire to stand up data centers as quickly as possible, but they need not represent the future.

Recent modeling shows off-grid solar microgrids could supply 90% of a datacenter’s power from renewables at roughly the same cost as Three Mile Island’s restart, while still maintaining small gas backup for around-the-clock reliability.¹⁶ This model reduces CO₂ emissions from 2.2 million tons to 0.3 million tons per year for a 500 MW capacity data center — representing a nearly 90% reduction, and reducing local pollution.¹⁷

If interconnection were more feasible from a regulatory angle, this argument could hold for grid-connected, distributed generation. Hyperscalers have a tremendous spending appetite, pouring investment through equity and power purchase agreements (PPAs) into novel green high power-density generation technologies, such as small modular nuclear reactors (SMRs).

Notably, utilities gain little from innovation and often bear disproportionate, consumer-shouldered risk from failed experiments—reflected by the industry’s chronically low R&D expenditure. Hyperscalers are emerging as well-incentivized and financially equipped catalyzers for technological breakthroughs.¹⁸

For example, in 2024, Amazon led a $500 million investment in X-energy, a nuclear reactor company, and committed to enabling over 5 gigawatts of new SMR capacity by 2039. Similarly, Google signed agreements for up to 500MW from Kairos Power’s reactors — a relationship that has now grown to include the Tennessee Valley Authority utility company, which will deliver that power to Google through its grid.¹⁹

“We have seen a remarkable surge in investments from hyperscalers into advanced energy technologies, which is still only a small fraction of their larger data center build-out,” says Charles Yang of the Center for Industrial Strategy. “Congress needs to strike a permitting deal, frankly by mid-2026, to ensure we can actually build and deploy the net-new generation capacity the country requires in a reasonable timeframe.” Public financing — in particular, grant and loan authorities at the Department of Energy — are essential for long-term stability of these projects, he added.

Modern data centers can also house ‘zero-water’ systems supporting continuous recycling. The data center boom has spurred cooling innovations across the board: Microsoft’s new data center designs pledge no evaporative losses, Amazon’s systems are completely closed loop, and NVIDIA’s Blackwell GPUs promise 300x improved water efficiency.²⁰

As a result of these modernizations, North America’s projected per-kWh cost for processing power is expected to decrease in the next five years.²¹ Energy generation should grow cleaner; water use will become more efficient. These are examples of hyperscalers indirectly investing billions of fresh R&D dollars into the grid—something utilities companies are structurally disincentivized to do.²²

While much NIMBYism raises valid concerns, opposition movements occasionally disregard how corporate stakeholders are aligned with local interests. Eindhoven — a city of under 250,000 people in the Netherlands — is home to ASML, the world’s leading lithography company and one of Europe’s key stakeholders in AI. In 2015, the company paid 77 million euros in Dutch taxes and opened almost 40,000 national jobs.²³ Even in Eindhoven, local newspapers and graffitied street corners condemn ASML’s existence and expansion.

Similarly, despite resistance, data center buildout in the U.S. is notably benefiting local communities. In Loudoun County, Virginia, roughly 200 data centers generate a third of the county budget while occupying just 3% of the land.²⁴ The resulting tax revenue funds nearby schools and has resulted in a massive local economic boom.

Related environmental or zoning concerns are legitimate — but they are also solvable.

AI Infrastructure “Flywheel”

America has been here before. In the 19th century, the transcontinental railroad buildout was originally justified on narrow commercial grounds: carrying goods more cheaply or connecting factories to markets. It was financed through a blend of public land grants and private investment and ultimately created a durable public benefit. What began as an industrial necessity stitching together communities became the backbone of American mobility and prosperity for that era. 

An AI-driven grid modernization presents a similar opportunity. What hyperscalers and utilities may first see as business infrastructure — moving electrons to feed clusters of servers — could become the scaffolding of a more resilient and innovative power system. As the railroads did, energy investments can transcend their initial purpose and become a fount of innovation and human capital development, too.

Cementing a first-mover advantage in AI leadership may be one of the greatest economic and scientific opportunities for Americans in recent memory. “The domestic energy demands of AI are enormous and only will increase in the future—grid modernization and energy efficiency therefore are central to our national competitiveness,” says Professor Ted Wittenstein of the Yale Schmidt Program on AI. “The United States must continue to lead in this crucial component of the AI stack, and we cannot cede our comparative advantage in domestic energy production and grid distribution.”

To achieve this aim, America must develop an incentive-aligned, compounding flywheel — for consumers and companies alike — across the entire AI stack. This naturally begins with the physical infrastructure that powers it all. And that falls upon the hyperscalers building this infrastructure, and the regulators supporting them.

There is a disconnect between what corporations can contribute to their communities, and locals’ perception of them. Hyperscalers, policymakers, and activists must collaborate to close this gap.

By first proactively addressing environmental concerns, bringing online cost-effective new generation, and investing private dollars in public infrastructure, hyperscalers can accentuate their mutual interests with the public — and act on them. With abundant generation capabilities, revamped interconnection infrastructure, and a more competitive supply-side market, consumers could benefit from a modernized grid where energy is cheaper, more reliable, and — through new projects unifying existing stranded green energy sources — potentially even cleaner.

This buildout is happening, but it can happen with headwinds or tailwinds. The more public support, the better off hyperscaler initiatives, and America’s continued ability to lead in AI, will be.

Quotes, anecdotes, and other specific details come from the authors’ personal notes and recordings from numerous conversations in South Korea, the Netherlands and San Francisco, CA with industry experts, engineers, and academics.

The authors thank Professors Ted Wittenstein and Phil Kaplan; and Benjamin Della Rocca, Charles Yang, Ajey Pandey, Anson Yu, Miquel Vila, Mark Ellis, Katherine He, and Mohit Agarwal for their thoughts, editorial feedback and support.

1. Holland, Steve. 2025. “Trump Announces Private-sector $500 Billion Investment in AI Infrastructure.” Reuters, January 22, 2025; Institute for Progress. 2025. “Compute in America: Building the Next Generation of AI Infrastructure at Home | IFP.” February 25, 2025. ↩︎
2. The average U.S. household uses about 900 kilowatt-hours per month—roughly 10,800 kWh per year. Even a single large-scale AI training run can draw power comparable to an entire city’s residential consumption. ↩︎
3. America’s energy grid is the United States ‘largest machine’, comprised of four primary entities: generation, transmission, distribution, and coordination. It’s enveloped by a rapidly moving commodity market and countless private power systems, referred to as “behind-the-meter” projects. ↩︎
4. Zeitlin, Matthew. 2025. “Heatmap Poll: Only 44% of Americans Would Welcome a Data Center Nearby.” Heatmap News, September 11, 2025. ↩︎
5. “Residents Cheer as Tucson Rejects Amazon’s Massive Project Blue Data Center Campus in Arizona.” 2025. DCD. August 8, 2025. ↩︎
6. Robinson, Dan. 2025. “NIMBYs threaten to sink Project Sail, a $17B datacenter development in Georgia.” The Register, August 22, 2025; “Mooresville Residents Reject North Carolina Data Center Project.” 2025. ↩︎
7. Ontiveros, Jeremie Eliahou. 2025. “Ajey Pandey.” SemiAnalysis. June 25, 2025. ↩︎
8. “South Korea to Pour 73 Trillion Won Into Power Infrastructure.” 2025. The Chosun Daily. May 28, 2025. ↩︎
9. “Amazon, Google, Microsoft Want to Build Own Power Lines in Ireland – Report.” 2024. DCD. June 10, 2024; Vincent, David ChernicoffMatt. n.d. “Google and CTC Global Partner to Fast-Track U.S. Power Grid Upgrades.” Data Center Frontier. ↩︎
10. Today, utilities are simply acting in their economic interest—and that of their shareholders, who are risk-averse investors seeking steady returns; these remain largely unaligned with those of hyperscalers and everyday Americans. ↩︎
11. For example, PJM Interconnection – one of the largest regional transmission organizations will charge consumers $14.7 billion from capacity in 2025-2026, according to utilitydive.com. This applies primarily to generators in competitive markets. Throughout this piece, “utilities” refers to both traditional cost-of-service-regulated utilities and unregulated generators in competitive electricity markets, unless otherwise specified. ↩︎
12. In fact, utilities companies have a substantially higher rate of return than their cost of capital, which means they’re providing notably better than market rate returns. Legal experts have highlighted this as a perverse structure, since the rate of return should expectedly equal the cost of capital in traditional capitalism. ↩︎
13. Additionally, when utilities rapidly rebuild capacity, they’re building their way out of future buildout projects, a prospect that does not thrill their shareholders. Even worse is re-conductoring existing lines, which is 75% cheaper and significantly faster than building net-new ones. ↩︎
14. High level is that adding high-load factor load increases the denominator (kWh) more than the numerator (cost), and therefore should reduce average rates over time. The wrinkle is that capital cost inflation may have outpaced general inflation (so it’s said, but you’ll want to confirm), so the higher cost of new capex may outweigh this effect. ↩︎
15. “A billionaire, an AI supercomputer, toxic emissions and a Memphis community that did nothing wrong.” n.d. TENNESSEE LOOKOUT. ↩︎
16. A collaboration between Scale Microgrids, Paces, and Stripe. ↩︎
17. Ibid. Although the One Big Beautiful Bill accelerates the phase-out of solar tax incentives, adds stricter sourcing and construction deadlines, and is forecast to reduce solar installations by about 15-20% over the next decade, it leaves a window open for many projects to qualify under the updated tax-credit rules if they begin construction soon. ↩︎
18. “Return on Equity.” n.d. ↩︎
19. Energy, X. 2024. “Amazon Invests in X-energy to Support Advanced Small Modular Nuclear Reactors and Expand Carbon-Free Power — X-energy.” X-Energy. October 16, 2024; Kaufman, Alexander C. 2025. “Google, Kairos, and TVA Ink Historic Next-generation Nuclear Deal.” Latitude Media, August 20, 2025; Terrell, Michael. 2024. “New Nuclear Clean Energy Agreement With Kairos Power.” Google, October 14, 2024. ↩︎
20. Solomon, Steve. 2025. “Sustainable by Design: Next-generation Datacenters Consume Zero Water for Cooling.” The Microsoft Cloud Blog. August 18, 2025; Davies, Alex. 2025. “AWS Rolls Out Liquid Cooling in Data Centers.” June 11, 2025; Rice, Sophie. n.d. “How Nvidia Boosts Data Centre Innovation With AI Cooling.” AI Magazine. ↩︎
21. Intelligence, Global Water. 2025. “New Data From Global Water Intelligence Reveals Impact of Hyperscale Data Center Boom on Onsite Water Consumption and Unprecedented Growth in Water Technology and Infrastructure Spending.” GlobeNewswire News Room, July 24, 2025; Nicoletti, Leonardo, Michelle Ma, and Dina Bass. 2025. “How AI Demand Is Draining Local Water Supplies.” Bloomberg.Com, May 8, 2025. ↩︎
22. As a contrast, utilities inject little-to-no R&D spend into the grid, given their revenue model is largely reliant on heavy capex investments with high return profiles. This also means opex cost shaving through tech modernization programs are rarely adopted by utilities companies. ↩︎
23. Danendra, Akira Farhan. 2024. “ASML Exodus: A Trouble in Holland’s Paradise.” Modern Diplomacy. June 17, 2024. ↩︎
24. Shamlian, Janet, and Alicia Hastey. 2025. “A Virginia County’s Economy Depends on Data Centers. Some Say They’re Intruding on Communities.” CBS News. February 20, 2025. ↩︎