Big Tech’s $690 Billion Spending Spree Could Strain the US Power Grid

Last updated 2 months ago. Our resources are updated regularly but please keep in mind that links, programs, policies, and contact information do change.

Five American technology companies plan to spend nearly $690 billion on infrastructure in 2026—almost double what they spent in 2025. Amazon alone is committing $200 billion. Microsoft, $120 billion. Google’s parent company Alphabet, up to $185 billion. Meta and Oracle are piling on another $185 billion combined.

But the U.S. doesn’t have enough electricity to power these facilities.

Microsoft’s CFO confirmed on the company’s most recent earnings call that Microsoft can’t fulfill $80 billion in customer orders for its computing services—not because of chip shortages or engineering problems, but because there isn’t enough electricity to run the equipment. CEO Satya Nadella stated: “The biggest issue we are now having is not a compute glut, but it’s power. If you can’t do that, you may have a bunch of chips sitting in inventory that I can’t plug in. In fact, that is my problem today.”

The question is who pays for the massive buildout required, what happens to everyone else competing for scarce electricity in the meantime, and whether your power bill is about to go up to subsidize AI facilities you’ll never use.

Tech Spending Projections vs. Grid Reality

Amazon’s $200 billion spending plan for 2026 exceeded analyst predictions of $147 billion. Alphabet revised its guidance to $175–185 billion. Microsoft is tracking toward $120 billion or more—it spent $37.5 billion in one quarter. Meta plans $115–135 billion. Oracle, $50 billion, representing a 136 percent increase over 2025.

The combined spending of these five firms exceeds the entire federal budget for border security and disaster response combined.

Yet investors aren’t celebrating. Amazon’s stock dropped 8–10 percent after announcing its spending plan, despite reporting strong cloud revenue growth and a 40 percent year-over-year increase in AWS order backlog. The stock decline reflects skepticism about whether these companies can actually build what they’re promising given physical constraints.

Microsoft’s experience illustrates this concern. The company spent tens of billions on NVIDIA GPUs that are now, according to leadership, “literally collecting dust” in warehouses because there isn’t enough electrical capacity to run them.

Solving a shortage requires cooperation of utility operators, regulatory agencies, and physical construction of new generation capacity—a process that takes 5 to 10 years from planning to operation.

The 175-Gigawatt Shortfall

By 2033, the United States faces a shortage of 175 gigawatts—enough to power 130 million homes.

The problem is that AI facilities are outbidding everyone else for access to available capacity. Tech companies have driven up electricity costs by $23.1 billion in the nation’s largest grid over three consecutive auctions. Technology firms now account for nearly half of all costs of buying electricity capacity.

This means they’re outbidding manufacturing plants, agricultural operations, and residential customers for scarce electrical capacity. These costs eventually flow through to everyone’s electricity bill.

Utilities predicted electricity demand would jump 237 percent, from 38 gigawatts in 2023 to 128 gigawatts in 2024, with AI facilities accounting for most of this revision.

In Texas, the grid operator projects that the safety buffer of extra power capacity will decline to negative values by 2026 or 2027. Negative reserve margins mean the grid wouldn’t have enough capacity to meet peak demand while maintaining the reliability buffer required by federal standards.

The Eastern grid operated by PJM Interconnection, which serves the mid-Atlantic region from New Jersey to Illinois, faces similar strain. PJM expects summer electricity demand to jump 42 percent—a 70-gigawatt increase over the next 15 years from the historical peak.

The Timeline Problem

Facilities can move from initial construction to operational in 18–24 months. New plants—whether solar farms, gas plants, or nuclear facilities—require 7–12 years from initial planning to delivering electricity to the grid.

Even with unlimited capital, new generation capacity cannot be built quickly enough to match demand growth.

Utilities have placed orders for 32 gigawatts of large gas turbines since 2024, the highest level in 20 years. But gas turbines won’t come online fast enough to match the acceleration in construction.

Regional concentration of AI operations makes this worse. Major tech firms are clustering facilities in Northern Virginia, parts of Texas, Ohio, and California—regions that already face limits on power delivery and production. Northern Virginia is experiencing such acute shortages that utilities are turning away new projects due to insufficient available capacity.

Big companies that locked in power contracts will win; smaller competitors will lose. Firms that secured capacity through long-term contracts will thrive. Competitors unable to access electricity face fundamental constraints on their AI scaling.

Power Consumption of AI Facilities

A major AI facility needs 200 to 500 megawatts of continuous electricity—equivalent to the supply for a city of 150,000 to 375,000 people running 24 hours a day, 365 days a year.

The densest configurations can exceed 1 gigawatt. One proposed facility in Indiana will require 2.4 gigawatts of dedicated electricity, with an additional 600 megawatts reserved to maintain grid reliability for everyone else in that region.

This demand doesn’t spike during the day and drop at night. These operations run around the clock.

NVIDIA’s upcoming Rubin Ultra NVL576 rack, expected in 2027, could consume up to 600 kilowatts per rack. A single facility with tens of thousands of racks could easily consume 500–800 megawatts of continuous electricity.

When too much demand concentrates in one area, power lines get completely full. A Bloomberg analysis found that wholesale prices more than doubled in some regions between 2020 and 2025, with many of the largest increases concentrated within 50 miles of major hubs.

New Power Plant Construction Timelines

AI data centers currently use about 4.4 percent of all U.S. electricity. By 2028, according to Department of Energy projections, that figure may reach 12 percent. This tripling of consumption must occur while the United States simultaneously electrifies transportation, electrifies heating in buildings, and expands electricity access for industrial applications.

Solar will supply the largest increase in generation, with approximately 70 gigawatts of new solar capacity scheduled to come online in 2026 and 2027—a 49 percent increase in U.S. solar operating capacity compared with the end of 2025.

Solar generation varies by time of day and weather conditions. To handle solar’s intermittency, utilities need batteries alongside solar installations. Texas plans to expand battery capacity from about 15 gigawatts in 2025 to 37 gigawatts by the end of 2027.

Natural gas will play a more immediate role. Companies are rushing to build natural gas power plants as a quick fix. Siemens Energy increased the number of gas turbines sold globally in fiscal 2025 to 194 units from 100 units in 2024, a 94 percent increase, with most driven by U.S. demand.

It takes 8-12 years to get permission to connect a new power plant to the grid and be physically connected.

Tech Companies’ Nuclear Investments

Facing timeline constraints, tech companies are betting on nuclear power. Amazon, Google, Meta, and Microsoft have announced major investments in nuclear to provide reliable, carbon-free power for AI operations.

Microsoft signed a $16 billion agreement to restart the shuttered Three Mile Island Unit 1 nuclear reactor in Pennsylvania, with the entire 835-megawatt output committed to powering Microsoft’s Azure operations. The reactor is expected to reopen in 2028.

Google signed an agreement with Kairos Power to develop a fleet of up to 500 megawatts of small modular reactor capacity, with reactors coming online through 2035. Amazon announced over $20 billion in investment to convert the Susquehanna nuclear site into an AI-ready campus powered entirely by nuclear energy.

These nuclear projects won’t be ready until 2030 or later. The first Kairos SMR is targeted for 2030, well after the acute 2026–2028 capacity crisis. For the immediate future, Big Tech must rely on natural gas plants and batteries.

Federal Regulatory Response

Recognizing that traditional interconnection processes can’t keep pace with demand, federal regulators are trying to speed up approval processes. On January 20, 2026, federal agencies announced an accelerated interconnection plan aimed at compressing decade-long timelines.

Federal energy regulators have been ordered to approve new power connections faster, by April 30, 2026—a deadline critics say is impossible to meet given the complexity of grid integration decisions.

FERC wants to let data centers build right next to power plants at or near generation facilities, so they can get power directly instead of through congested power lines. Amazon’s agreements with Talen Energy to locate facilities at nuclear plants exemplify this model.

This helps the grid but creates fairness problems for other customers. What happens if a co-located facility experiences equipment failure? Who gets priority when electricity is scarce?

The Trump administration is using executive orders to speed things up. Executive orders have told federal agencies to cut red tape, reduce permitting timelines, streamline environmental review processes, and use emergency declarations to expedite construction.

No amount of paperwork can speed up building power plants. Building power lines and plants still takes years. A federal regulator warned that the grid can’t handle this without major upgrades: “without sufficient amount of grid and reliability upgrades to the system, we can’t operate.”

Regional Impacts: Texas, Mid-Atlantic, and Stargate

The power shortage will hit some regions much harder than others due to both the concentration of planned construction and pre-existing capacity constraints.

Texas is the epicenter. The Stargate project—a massive AI facility partnership—is building across Texas, a $500 billion joint venture between OpenAI, SoftBank, and Oracle. Initial capacity from Stargate is targeted to reach 7 gigawatts by late 2025, with additional sites in Shackelford County, Doña Ana County in New Mexico, and Ohio.

ERCOT’s December 2024 report projected that reserve margins for summer 2026 would be 4.5 percent—already below the required safety level of 13.75 percent—meaning the grid won’t have enough power to meet demand by 2027–2029.

PJM, which operates the power grid for the mid-Atlantic region from New Jersey to Illinois, faces similarly acute challenges. PJM’s 2025 forecast projected that summer peak usage will increase by approximately 70 gigawatts over the next 15 years, with the bulk of this growth concentrated in 2025–2030 as facilities come online. Electricity demand is projected to jump 30 gigawatts between 2025 and 2030—roughly 18 percent growth in PJM’s total peak demand.

California projects more modest load growth by comparison—only 1.8 gigawatts by 2030 and 4.9 gigawatts by 2040. But California relies on solar, which doesn’t work in the evening when demand peaks.

Water Consumption and Competition

These facilities require vast quantities of water for cooling. A mid-sized facility uses millions of gallons of water per day, while larger operations require up to 5 million gallons per day. U.S. water use for these facilities could double or quadruple by 2028 to roughly 150–280 billion liters per year.

This demand is concentrated in water-stressed regions. Two-thirds of new facilities are in areas that already don’t have enough water, including hot, dry climates like Arizona. A Meta facility that opened in 2018 in Newton County, Georgia, consumes 500,000 gallons of water per day—10 percent of the entire county’s water consumption.

Tech companies, towns, and farmers are competing for the same water. Some utilities have refused to approve new facilities due to water shortages.

Cost Distribution and Rate Impacts

For ordinary Americans, this means your electricity bill could go up. Electricity prices are already rising in areas with lots of data centers, and these costs show up on your electricity bill.

People living near data centers could see significant increases in their power bills as utilities try to pay for new power plants and power lines driven primarily by technology firm demand.

The question of whether tech companies should pay for all the upgrades they need, or whether these costs should be spread across all ratepayers, has become politically contentious. Policymakers from both parties have signaled that they expect operators to pay, but there’s no agreement on how to make that happen.

Some states are requiring tech companies to sign long-term contracts or put down deposits, but there’s no national standard yet.

Demand Management and Efficiency Solutions

Instead of building new power plants, there are other ways to manage the capacity crisis without waiting a decade for new generation construction. These approaches reduce how much power is needed at peak times rather than building only new supply.

Demand response means turning down power use when the grid is stressed. AI training can be paused during peak hours without losing much progress, unlike AI chatbots that need to answer questions instantly. Companies showed that AI workloads could cut power use by 25 percent by delaying less urgent tasks during peak hours. If this becomes standard, it could free up enough power for 126 gigawatts of new demand immediately.

Texas is requiring data centers to cut power use during peak hours, signaling that this will become mandatory to get grid access.

Solar panels plus batteries could be another solution. As batteries got cheaper, solar plus storage became cost-competitive with natural gas and can now cost less than building new gas plants in some markets. Battery storage is being deployed globally at record rates, with 106 gigawatts installed in 2025.

Making AI chips more efficient is another solution. Google cut the power needed to run its AI by 78 percent through better design over 2025, demonstrating that better software design can cut power use dramatically.

However, when efficiency improvements increase total usage instead of reducing it, cheaper AI encourages more AI use, which cancels out the savings.

Critical Milestones in 2026 and Beyond

2026 is when things will get real. By mid-2026, the first wave of tech company spending should be under heavy construction, with some opening. We’ll see if the power demand matches the predictions as these operations begin.

If demand is as high as expected, electricity prices will spike in tech hubs, forcing politicians to act.

FERC’s April 30, 2026 deadline for new rules on connecting data centers will set the rules for who gets power and who pays for upgrades for years to come. The rules will decide if data centers can tap power plants directly, whether they pay for power line upgrades, and if your bill goes up for residential and small business ratepayers.

The Stargate project is supposed to be partly operational by late 2025, with 7 gigawatts of capacity, with more coming online through 2026 and beyond. How fast these get built will show if federal streamlining works.

Three Mile Island is expected to open in 2028, but only after the crisis hits, providing partial relief. Whether this nuclear plant works will show if tech companies’ nuclear bets can solve the power problem long-term.

Three factors will determine if this works: whether faster approvals speed up building power lines; whether cutting peak demand, batteries, and better chips reduce power needs; and whether nuclear plants and new power sources open in time to prevent blackouts.

The outcome will determine how fast AI grows, but also whether the power grid stays stable and affordable for everyone who depends on it.

Our articles make government information more accessible. Please consult a qualified professional for financial, legal, or health advice specific to your circumstances.