Data as of November 12, 2026
In Brief
- Natural gas is the largest energy source for data centers in the United States. Solar is the fastest-growing, while nuclear is considered the ideal long-term option.
- Oil is rarely used for electricity generation in the U.S., so its correlation with AI-driven energy costs is weak.
- Google acquired an entire power developer because the grid cannot connect new data centers fast enough.
- Capital expenditures (CapEx) for Big Tech companies continue to rise even as free cash flow shrinks.
Why Do AI Data Centers Consume So Much Power?
A single advanced AI chip consumes between 700 and 1,000 watts, while a single AI server rack can draw as much power as dozens of homes. Cooling alone can account for 38% to 40% of total electricity usage in data centers. The International Energy Agency (IEA) projects that global data center electricity demand will more than double to roughly 945 terawatt-hours (TWh) by 2030, with the United States taking the largest share—representing nearly half of U.S. demand growth.
Google’s electricity consumption rose from 31 TWh in 2024 to 43 TWh in 2025, with its data centers alone accounting for 42.4 TWh. This exceeds the 2024 total electricity consumption of states like Nevada (40.7 TWh), Utah (34.7 TWh), Connecticut (27.4 TWh), and Idaho (26.4 TWh), though it remains below Iowa (54.9 TWh) and Oregon (59.7 TWh). Note that Google’s figures are global for 2025, whereas state figures are U.S.-specific for 2024.
Which Energy Source Is Used the Most?
The U.S. Power Grid in 2025: Natural gas contributed about 41%, nuclear 18%, coal 17%, and renewables 24%, while petroleum did not exceed 1%. Solar power accounted for roughly 7% of utility-scale electricity generation.
Natural Gas: According to the IEA, natural gas meets 40% of U.S. data center power needs and will likely fuel most capacity expansions through 2030. Industry insiders describe it as the fastest, cheapest, and easiest option for many companies. The U.S. Energy Information Administration (EIA) forecasts natural gas power generation to grow by 23% in Texas’s ERCOT grid and 5% in the PJM interconnection between 2025 and 2027, driven partly by data centers.
Oil: Petroleum contributed only about 0.7% of U.S. electricity in 2025. The 2026 Iran conflict tested this dynamic: global oil prices surged 50% between February and May, while U.S. natural gas prices fell by 6%. Oil still exerts an indirect influence via inflation and interest rates, though no direct link has been established between oil prices and AI stocks.
Solar Power: Wind and solar combined generated a record 17% of U.S. electricity in 2025, with utility-scale solar growing by 34%. The EIA identifies solar as the fastest-growing U.S. power source through 2027. In February 2026, Google signed a deal with TotalEnergies to supply 1 GW of solar power to its Texas data centers. However, because solar only generates power when the sun shines, it requires battery storage or backup systems.
Nuclear Power: Microsoft’s Three Mile Island unit (835 MW) is scheduled to come online in 2027–2028, while Meta contracted for 1,121 MW from Constellation’s Clinton plant and 2,176 MW from three Vistra plants. Some of these agreements purchase output from existing reactors rather than adding new supply to the grid.
Why Are Tech Giants Acquiring Power Companies?
Alphabet signed a deal to acquire Intersect Power on December 22, 2025, for $4.75 billion in cash plus assumed debt, closing the transaction on March 10, 2026. This marked the first major tech acquisition of a renewable energy developer, shifting Google from buying power via contracts to owning power generation assets directly. The deal also secures projects that already hold spots in grid interconnection queues.
Key reasons cited by industry sources include:
- Grid Delays: Data centers can be built within 18 to 24 months, prompting buyers to acquire power developers to bypass grid bottlenecks and secure supply.
- Co-location: Company-owned power generators and batteries can be built directly alongside data centers in dedicated "energy parks."
- Political Pressure: A White House pledge urges tech firms to supply, build, or procure their own power for data centers—signed by Google, Amazon, Meta, Microsoft, Oracle, OpenAI, and xAI.
- Varying Tactics: Amazon, by contrast, has opted to purchase individual stalled energy projects rather than acquiring an entire developer.
Owning power infrastructure increases capital and operational risks. Alphabet’s free cash flow turned negative in Q2 2026, with CreditSights estimating that CapEx will consume 46% of its revenue.
The Competition and Where It Is Headed
The four largest hyperscalers plan to invest roughly $720 billion to $745 billion in CapEx in 2026. Amazon, Microsoft, Google, and Meta accounted for 49% of global corporate clean energy procurement volumes in 2025. The IEA’s base-case scenario projects data center power demand reaching around 1,200 TWh by 2035, within a range of 700 to 1,700 TWh depending on chip efficiency and grid constraints.
Notes
CapEx projections shift with every quarterly earnings report. Certain deal figures are drawn from specialized industry trackers cited in the text. Please verify details against official company filings before relying on any specific metric.
Sources
- U.S. Energy Information Administration (EIA): Electricity in the United States
- EIA: Wind and solar generated a record 17% of U.S. electricity in 2025
- EIA: Solar power generation drives electricity generation growth
- International Energy Agency (IEA): Energy and AI, Executive Summary
- The Irish Times: AI data centre growth means more coal and gas plants, IEA says
- Latestly: Why the cloud still runs on coal and gas
- Congressional Research Service (CRS): Effects of Iran Conflict on Natural Gas Prices
- Google 2026 Environmental Report
- PV Tech: Google finalises Intersect acquisition to bring data centre power in-house
- Introl: Google Intersect Power Acquisition & Energy Parks 2026
- The Brief Script: AI Data Center Acquisitions & Investment 2026
- EE Times: Google Bets Big on Power Generation to Fuel AI
- Propmodo: Power Acquisitions Surge as Data Center Developers Bypass Grid Delays
- Axis Intelligence: Big Tech & Clean Energy 2026
