Why Tech Giants Are Now Buying Power Plants

The Corporate Data Center Land Grab: Why Tech Giants Are Buying Power Plants, Not Just Chips

Amazon confirmed in August 2026 that it has acquired an 8,000-acre site in Pecos County, Texas, where it plans to build a data center powered by what could become the largest natural gas power plant in the United States — permitted for 35 turbines generating up to 7.65 gigawatts, enough for millions of homes, and deliberately disconnected from the Texas grid. It's not an isolated bet. Amazon joins Microsoft, Google, and Meta, all of which have committed to building or restarting power plants of their own in the past 18 months rather than simply buying electricity from a utility. The reason isn't a sudden interest in energy as a business — it's that the U.S. grid can no longer connect new power fast enough to keep pace with AI buildout, and interconnection queues now run five to eight years. Chips can be delivered in months. Grid connections can't. So the hyperscalers are increasingly skipping the grid altogether.

The scale of what's actually being built

This isn't a handful of pilot projects. According to research firm Cleanview, cited in reporting on the Amazon deal, data center developers have announced nearly 60 behind-the-meter gas power projects since the start of 2025, with a combined capacity of roughly 90 gigawatts. Some of the individual numbers are startling on their own: Amazon's Pecos County site is permitted for up to 7.65 GW of on-site gas generation, according to permit filings reviewed by Cleanview and reported by Data Center Dynamics. Microsoft is working with Chevron and Engine No. 1 on an off-grid gas plant in West Texas designed to scale to 5 GW, TechCrunch reported in April 2026. Google confirmed a partnership with Crusoe Energy to build a 933-megawatt gas plant tied to its Goodnight data center campus in the Texas panhandle, according to Yale Environment 360 — the third gas project Google has been tied to in recent months, following an October 2025 deal for power from a plant in Illinois. Meta added seven more gas power plants to its Hyperion data center site in Louisiana, bringing that single campus to 7.46 GW of capacity — described by TechCrunch as enough to power the entire state of South Dakota.

On the nuclear side, the pattern is similar but slower-moving. Microsoft is paying $16 billion over 20 years to restart Three Mile Island's undamaged sister reactor, now renamed the Crane Clean Energy Center, expected back online in 2027, according to Forbes. Meta has struck deals for up to 6.6 GW of nuclear capacity through 2035, including a 20-year power purchase agreement covering more than 2.1 GW from Vistra's Beaver Valley plant in Pennsylvania and its Perry and Davis-Besse plants in Ohio, according to Meta's own disclosures cited in enCore Energy's 2026 SEC filing. Tracking site SMRintel puts total committed nuclear capacity for data centers across all four hyperscalers at roughly 9.8 GW as of mid-2026.

Why grid connections, not chips, are the real bottleneck

The move away from the public grid isn't about cost savings or environmental branding — it's about speed. Getting a new power source connected to the existing grid means entering an interconnection queue that, per industry reporting, can now stretch five to eight years in the regions where data centers are concentrated. A behind-the-meter gas plant, by contrast, can be deployed in as little as 18 months, according to Williams Companies' account of its own shift toward on-site generation, reported by energy-industry outlet EnkiAI.

That gap matters because a data center campus without power is just an empty building full of chips no one can run. Hyperscalers that once treated electricity procurement as a utility negotiation are now treating power generation as critical-path infrastructure they have to control directly — building or buying it themselves rather than waiting for someone else to build it for them.

The part nobody agrees on: what this does to household electricity bills

Whether this shift is actually raising consumer electricity prices is genuinely disputed, and the disagreement matters more than either side's press release suggests.

On one side, an Electric Power Research Institute study found that data centers put downward pressure on average electricity prices through 2024, and a March 2026 Institute for Energy Research study found no statistically significant correlation between data center concentration and current electricity prices, according to reporting on both studies by Marketplace and the Independent Institute.

On the other side, the evidence of localized price pain is becoming harder to dismiss as anecdotal. PJM, the largest grid operator in the country, projected in a report covered by Fortune that a $6.3 billion increase in consumer electricity costs over the next three years can be mostly attributed to rising data center power demand. In Virginia — the state with the heaviest data center concentration in the country — residential electricity prices rose more than 13% in the past year, according to U.S. Energy Information Administration data cited by Fortune. Zooming out further, PolitiFact reported in June 2026 that average residential electricity prices rose 94% in Washington, D.C., 74% in Maryland, and 73% in Maine between March 2021 and March 2026, per EIA figures, though data centers are only one of several drivers behind those increases. Goldman Sachs, for its part, has projected the AI buildout will raise electricity costs 6% between 2026 and 2027 and another 3% by 2028.

The honest read is that the two sets of findings aren't necessarily contradictory — data centers may have modestly lowered average prices while demand still exceeded supply, and now be pushing prices up sharply in the specific regions where new generation hasn't kept pace with new load. That's precisely the dynamic that's pushing hyperscalers toward off-grid generation in the first place: building their own supply means a company's AI buildout doesn't have to compete with residential ratepayers for the same limited pool of new power.

What this actually signals

The framing that's stuck in tech coverage — that this is an AI story, or a chips story — misses what's actually happening on the ground. It's an energy infrastructure story, and the companies driving it are behaving less like software companies and more like independent power producers who happen to also run data centers.

We believe in paying the full costs of powering our operations.

Amazon, company statement on the Pecos County project

That's a pointed way of pre-empting the criticism that its buildout is what's driving up bills for everyone else on the grid.

Whether that pledge holds up matters beyond any one company's reputation. If off-grid generation genuinely insulates AI buildout from the public grid, it could ease the pressure on residential rates that Fortune, Brookings, and PJM have all flagged. If the boundary between "our own power" and "grid-connected power" turns out to be leakier than the press releases suggest — through shared transmission infrastructure, water use, or local air permits — the political backlash already building in communities that have blocked or delayed roughly $98 billion in data center projects, according to research firm Data Center Watch, is unlikely to fade.