AI "competes with the public for electricity" in the U.S., nuclear power becomes Silicon Valley's "last hope"
TechFlow Selected TechFlow Selected
AI "competes with the public for electricity" in the U.S., nuclear power becomes Silicon Valley's "last hope"
Data centers are being built too fast, while grid expansion is too slow.
Navigating Web3 tides with focused insightsU.S. AI companies are once again busy investing in power plants.
Recently, Meta signed a long-term power purchase agreement with U.S. energy company Vistra to directly procure electricity from several of its operational nuclear power stations. Earlier, Meta also partnered with advanced nuclear firms Oklo and Terra Power to advance the commercial deployment of small modular reactors (SMRs) and fourth-generation nuclear technologies.
According to disclosures by Meta, if these collaborations proceed as planned, by 2035, Meta could secure up to approximately 6.6 GW (gigawatts, 1 GW = 1,000 MW/megawatts = 1 billion watts) of nuclear power supply.
Over the past year, large-scale energy investments by North American AI firms have become increasingly common: Microsoft is pushing for the restart of retired nuclear plants; Amazon is deploying data centers around nuclear stations; Google and xAI continue to ramp up long-term power purchase agreements. Amid intensifying competition in computing power, electricity is shifting from a cost factor to a strategic resource that AI companies must secure well in advance.
On the other hand, surging energy demand driven by the AI industry continues to strain the U.S. power grid.
According to foreign media reports, driven by soaring AI demand, PJM Interconnection—the largest grid operator in the United States—is facing severe supply-demand challenges. This electricity network, covering 13 states and serving about 67 million people, is approaching its operational limits.
PJM forecasts that electricity demand will grow at an average annual rate of 4.8% over the next decade, with nearly all new load coming from data centers and AI applications, while power generation and transmission infrastructure development clearly lags behind this pace.
The International Energy Agency (IEA) predicts that AI has become the most significant driver of rising electricity consumption in data centers, and expects global data center electricity use to reach approximately 945 TWh by 2030—doubling current levels.
The real mismatch lies here: AI data centers typically take only 1–2 years to build, whereas constructing a new high-voltage transmission line often takes 5–10 years to complete. In this context, AI companies are stepping directly into the arena, launching an unconventional wave of "mega infrastructure" projects involving investing in and building power plants.
01 AI Giants Rushing to Build Nuclear Power Plants
For over a decade, AI companies’ main activity on the energy front was “buying” rather than “producing” power—securing wind, solar, and some geothermal electricity through long-term power purchase agreements to lock in prices and meet carbon reduction goals.
Take Google as an example: the tech and AI giant has signed tens of gigawatts' worth of long-term renewable energy contracts globally and collaborates with geothermal firms to ensure stable clean power supply for its data centers.
In recent years, however, as AI-driven power consumption soars and grid bottlenecks emerge, some companies have begun participating in power plant construction or forming deep partnerships with nuclear stations—shifting their role from mere consumers to active participants in energy infrastructure.
One approach involves reviving retired power plants. In September 2024, Microsoft signed a 20-year power purchase agreement with nuclear operator Constellation Energy to support the restart of an 835-megawatt decommissioned nuclear unit, ensuring long-term power supply.
Joining Microsoft in this effort is the U.S. government: in November last year, the Department of Energy completed disbursement of a $1 billion loan to partially finance the project. The reactivated unit has been renamed the Crane Clean Energy Center (formerly Three Mile Island Unit 1).
In fact, Crane is not the only power station returning to service after retirement. In Pennsylvania, the Eddystone fossil fuel plant was originally scheduled to retire by the end of May 2024 but was urgently ordered by the Department of Energy to remain operational to prevent a power shortfall in the PJM region.
Meanwhile, Amazon’s cloud computing arm AWS took a different path—directly acquiring a data center adjacent to a nuclear power station. In 2024, energy firm Talen sold its roughly 960-megawatt data center campus near the Susquehanna nuclear plant in Pennsylvania to AWS. In June last year, Talen announced expanded cooperation, planning to supply up to 1,920 megawatts of carbon-free electricity to AWS data centers.
In terms of new power plant developments, Amazon has participated in the development of a small modular reactor (SMR) nuclear project in Washington State through investment and collaboration, led by organizations such as Energy Northwest. Each unit is about 80 megawatts, scalable to hundreds of megawatts overall, aiming to provide long-term, stable baseload power for data centers.
Google, for its part, partnered with U.S. nuclear company Kairos Power in 2024 to advance plans for new advanced nuclear reactor projects, targeting initial operations around 2030 and delivering approximately 500 megawatts of stable, carbon-free nuclear power by 2035 to support long-term data center operations.
Among these efforts, Meta stands out as one of the most aggressive players. To date, it has planned or secured access to up to 6.6 gigawatts of nuclear power resources. For comparison, the total installed capacity of currently operating nuclear plants in the United States is about 97 gigawatts.
All these initiatives fall under Meta’s “Meta Compute” framework—an overarching strategy unveiled earlier this year aimed at unifying future planning for computing and power infrastructure required by AI.
Data from the International Energy Agency shows that global data center electricity consumption will double by 2030, with AI being the primary driver. The United States accounts for the largest share of this growth, followed by China.
Meanwhile, previous projections by the U.S. Energy Information Administration (EIA) suggesting stable power generation capacity through 2035 have clearly been disrupted by the AI boom.
Based on publicly available information, by 2035, U.S. AI giants including Microsoft, Google, Meta, and AWS are expected to directly or indirectly secure more than 10 gigawatts of nuclear power capacity—and new infrastructure projects continue to be announced regularly.
AI is emerging as a new financial backbone for nuclear energy revival. On one hand, this reflects a pragmatic corporate choice—compared to wind and solar, nuclear offers stable 24/7 output, low-carbon emissions, and independence from large-scale energy storage; on the other, it aligns closely with supportive policy environments.
In May 2025, President Trump signed four executive orders on “nuclear energy revival,” proposing to quadruple U.S. nuclear capacity within 25 years and positioning it as part of national security and energy strategy.
Within the following year, stocks of nuclear-related companies rose significantly: operators like Vistra saw cumulative gains exceeding 1.5 times; SMR-focused firms such as Oklo and NuScale experienced even sharper increases, with stock prices multiplying several-fold.
Suddenly, propelled by AI industry investment and government backing, nuclear power has returned to the center of U.S. energy and industrial policy discussions.
02 Models Run Fast, But Power Plants Don’t Get Built Fast
Despite the boost in investor sentiment from the “nuclear revival,” nuclear power still accounts for only about 19% of U.S. electricity generation, and both new construction and plant restarts typically span a decade or more. In other words, the risk of AI overloading the power system remains unchanged.
In multiple long-term forecasts, PJM has warned that nearly all future load growth over the next ten years will come from data centers and AI applications. If power generation and transmission infrastructure cannot accelerate, grid reliability will face serious threats.
As one of the largest regional transmission organizations in the U.S., PJM covers 13 states and Washington D.C., serving around 67 million people—its stability directly affects core economic regions in the eastern and central United States.
While massive capital flows into power infrastructure, the issue of power congestion persists unresolved.
Underlying this contradiction is a severe mismatch between the rapid expansion of the U.S. AI industry and the slow pace of power system construction. Building a hyperscale AI data center usually takes just 1–2 years, whereas constructing new transmission lines and completing interconnection approvals can take 5–10 years.
As data centers and AI workloads keep consuming more electricity,新增 generating capacity fails to keep up. The ongoing strain on power resources leads directly to skyrocketing electricity prices.
In areas like Northern Virginia, where data centers are highly concentrated, residential electricity prices have surged dramatically in recent years, with some regions seeing increases exceeding 200%, far outpacing inflation.
Some market reports indicate that in the PJM region, as data center loads surge, costs in the power capacity market have risen sharply: The total capacity cost for the 2026–2027 auction period reached about $16.4 billion, with data center-related costs recently accounting for nearly half of the total. These increased costs are ultimately passed on to ordinary consumers through higher electricity bills.
As public frustration grows, power congestion has quickly escalated into a social issue. Regulatory bodies in states like New York have explicitly proposed requiring large data centers to bear greater responsibility for their surging electricity demand and associated grid connection and expansion costs—including higher connection fees and long-term capacity obligations.
"Before ChatGPT, we had never seen such a load increase," Tom Falcone, chairman of a major U.S. public power commission, said publicly. "This is a supply chain-wide problem involving utilities, industries, labor, and engineers—these aren't people you can just conjure out of thin air."
In November last year, PJM's market monitor filed a formal complaint with the Federal Energy Regulatory Commission (FERC), recommending that PJM halt approval of any new large data center interconnection projects until relevant procedures are improved, citing concerns over reliability and affordability.
To cope with the massive power demands of AI data centers, some U.S. states and utilities have started creating dedicated “data center tariff classes.” For instance, in November 2025, Kansas introduced new pricing rules requiring large power users (such as data centers) above 75 megawatts to enter long-term contracts and assume shared responsibilities for electricity pricing and infrastructure upgrade costs—ensuring they shoulder a larger portion of grid fees and modernization expenses.
Microsoft President Brad Smith recently stated in an interview that data center operators should “pay our way”—covering higher electricity rates or related charges for their usage, grid connections, and upgrades, rather than shifting costs onto regular electricity consumers.
Overseas, regions outside the U.S.—including Amsterdam, Dublin, and Singapore—have paused numerous new data center projects in recent years, primarily due to insufficient power infrastructure.
Under tighter constraints on power and land, data center expansion has become a stress test for a nation’s foundational infrastructure and capital mobilization capabilities. Beyond China and the U.S., few economies possess the engineering capacity to match such demands.
Indeed, the current power crunch in the U.S. makes clear: even massive investments in new power plants may not be enough to resolve the energy crisis of the AI era.
03 Need to Build Grids—and Also Look to the Skies
Beyond power plants, the deeper structural issue behind power congestion lies in the long-standing lag in U.S. transmission grid development.
Industry reports show that in 2024, the U.S. added only 322 miles of high-voltage transmission lines (345 kV and above)—one of the slowest construction years in the past 15 years, compared to nearly 4,000 miles added in 2013.
Insufficient transmission capacity means that even if more power plants come online, electricity may not be effectively delivered to densely populated demand centers due to limitations in long-distance transmission.
Between 2023 and 2024, PJM repeatedly warned that due to sluggish transmission development and inadequate generation resources, the growing load from new data centers has forced grid operators to adopt extraordinary measures to maintain system stability—including considering options such as cutting power to certain data centers or requiring them to operate on backup generators during peak demand periods, otherwise reliability risks would further escalate.
By contrast, China—often dubbed the “infrastructure superpower”—has consistently maintained high growth rates and technological iteration in grid construction. In recent years, China has aggressively expanded ultra-high voltage (UHV) transmission networks, commissioning multiple ±800 kV and 1,000 kV UHV lines between 2020 and 2024, adding thousands of kilometers of transmission lines annually.
In terms of installed capacity, China is projected to exceed 3,600+ gigawatts in total capacity in 2025, reflecting steady growth from 2024, with plans to add 200–300 gigawatts of renewable generation capacity throughout the year.
This gap in grid infrastructure capability cannot be easily closed by policy or capital alone in the short term.
Amid surging AI loads, the Federal Energy Regulatory Commission (FERC) formally issued Order No. 1920 in May 2024, finalizing its regional transmission planning reform initiated in 2021. The new rule requires utilities to conduct 20-year forward-looking planning and include cost-sharing discussions for new types of loads such as data centers.
However, given the lengthy timelines for regulation implementation, project approvals, and construction, this policy functions more as a mid-to-long-term “grid reinforcement” tool. Real-world pressure from power congestion will persist. In this context, space-based computing has emerged as a new frontier attracting industry attention.
In recent years, the global tech sector has advanced the concept of “space computing”—deploying computing nodes or data centers with AI training/inference capabilities in low Earth orbit (LEO) to overcome terrestrial limitations in energy, cooling, and connectivity.
SpaceX, for example, views low-orbit satellites and inter-satellite laser communications as the foundation for building a distributed “orbital computing network.” Leveraging its Starlink constellation, SpaceX is exploring onboard edge computing for remote sensing processing and real-time inference, reducing ground data return and energy burdens.
Meanwhile, startup Starcloud launched its Starcloud-1 satellite in November 2025, equipped with an NVIDIA H100 chip, successfully completing in-orbit inference validation. This case demonstrates that space-based computing is moving toward practical deployment.
China is also accelerating its space computing initiatives. The “Three-Body Computing Constellation,” led by Zhejiang Lab, has already launched its first batch of 12 satellites, with an official plan targeting a total computing capacity of 1,000 POPS, intended for orbital edge computing, massive data preprocessing, and AI inference.
Yet, whether space-based computing or next-generation energy systems, both remain in early validation stages. This explains why, over the past year, U.S. AI giants have raced to invest in nuclear power plants and other power infrastructure.
“We need clean, reliable sources of power that can run 24/7, every day of the week,” said IEA Executive Director Fatih Birol in a recent interview. “Nuclear energy is returning to center stage globally.”
Given that grid expansion and power plant construction cannot catch up in the short term, the current power crunch in the U.S. is unlikely to ease soon. Continuous large-scale capital investment in electricity—especially the nuclear sector—remains the only viable option for now.
Wood Mackenzie’s latest forecast indicates that as data centers and AI workloads continue driving up electricity demand, U.S. nuclear generation could rise approximately 27% above current levels after 2035.
According to foreign media reports, the U.S. government is supporting nuclear equipment manufacturers like Westinghouse through Department of Energy loans, export credits, and demonstration projects, promoting new reactor construction and life extensions of existing units to rebuild nuclear industrial capacity.
Against the dual drivers of industry momentum and policy support, U.S. AI giants will remain tightly coupled with the nuclear industry for the foreseeable future.
Join TechFlow official community to stay tuned
Telegram:https://t.me/TechFlowDaily
X (Twitter):https://x.com/TechFlowPost
X (Twitter) EN:https://x.com/BlockFlow_News
