The Data Center Backlash Is Now an Infrastructure Problem, Not Just a PR One

If you've been tracking the AI hardware buildout purely through GPU shipment numbers and MFU benchmarks, you're missing half the story. The physical infrastructure supporting all that compute, the power plants, generators, and grid connections, has become its own engineering and political mess. And the scale of what's getting built (and contested) is genuinely wild.

Start with the raw numbers. Nvidia just guaranteed up to $105 billion to back an OpenAI data center leased from SB Energy, a SoftBank subsidiary, in Ohio. The deal locks in 8 gigawatts of capacity, with the first 800 megawatts coming online in 2028. Nvidia's also kicking in a separate $1.5 billion investment directly into SB Energy. For context, a gigawatt is roughly what a large nuclear reactor produces, so this single facility is aiming for the output of eight reactors dedicated to one company's compute needs.

Amazon's playing a similar game in West Texas, but with a twist that should catch any infrastructure engineer's attention: it's investing in a gas-burning power plant that, at least initially, won't connect to the state grid at all. The GW Ranch facility in Pecos County will run 35 natural-gas turbines to deliver 7.65 gigawatts almost exclusively to Amazon's data center, according to the New York Times. Per Cleanview, which tracks these projects, the plant holds a Texas permit allowing emissions up to 33 million tons of CO2 annually, more than the largest coal plant in the country is permitted to emit. Plants rarely hit their max permitted output, but the fact that the ceiling is set that high tells you something about how loose the regulatory guardrails still are.

That's the pattern across the sector right now: build fast, secure power however you can, and worry about compliance later. A Microsoft-backed facility in New Jersey, the DataOne project in Vineland, has reportedly been running gas-fired generators without the permits developers are typically required to obtain before installing equipment that creates local air pollution, according to Floodlight. It's not an isolated case. Global Energy Monitor found that proposals for new gas-fired power capacity tied to U.S. data centers nearly doubled in the first half of 2026, with almost a third of that new capacity concentrated in Texas alone.

The Regulatory Rollback

Here's where it gets more consequential for anyone building or siting infrastructure long-term: the rules governing public input on this stuff are actively shrinking.

The EPA is planning to scrap a federal rule that requires public notice and comment periods when industrial sites, including data centers, apply for air permits. Advocates warn this would let developers break ground without giving nearby residents any warning, let alone a chance to object. It's a direct response to (or maybe an enabler of) the backlash that's been building nationwide.

That backlash is real and it's bipartisan, which is unusual enough to be notable on its own. The Washington Post has charted how online statements from both Democratic and Republican politicians shifted from mostly positive on data centers in early 2025 to mostly negative now. Texas Governor Greg Abbott, no one's idea of an anti-industry politician, told ABC's This Week that the sector moved too fast and "basically dug their own grave for the problem that's been caused for them." Pennsylvania Governor Josh Shapiro signed a wide-ranging executive order reining in data center development in his state. All candidates for Ohio governor have now called for restrictions on the boom, according to the Ohio Capital Journal.

At the local level, this is playing out as a wave of moratoriums and stalled permits:

• New York became the first state to enact a data center moratorium
• Louisville's Metro Council approved its own moratorium as a new proposal sparked fresh controversy
• Flagler County, Florida instituted a one-year construction ban
• Palm Beach County is considering following suit
• Austin's City Council is weighing new rules for large facilities
• A proposal in Lovejoy, Georgia was withdrawn outright after community pushback
• Taylor, Texas tabled a proposal for 90 days after residents pushed back

Meanwhile, Amazon employees in Seattle have asked the city to slow down on new data center approvals, and a data center developer is now suing an Oakland County, Michigan city over local restrictions. This isn't confined to rural areas anymore either. Loudoun County, Virginia, long the epicenter of US data center density, is dealing with its own resident backlash as the buildout intensifies.

There's also churn at the top of the industry's org charts worth flagging. Chris Malone, OpenAI's head of data centers and a key figure in the company's build-out plans, left last week after previously holding similar roles at Meta and Google. He'd reported directly to OpenAI president Greg Brockman until a reorg earlier this year shifted that reporting line. Departures like this, at a company racing to lock down gigawatts of capacity, are the kind of signal worth watching even if the public reasoning stays vague.

What to Watch

The core tension here isn't going away: AI compute demand keeps climbing, and the power infrastructure to support it takes years to permit and build under normal rules. Companies are responding by building off-grid, leaning on gas turbines with generous emissions ceilings, and, in at least one alleged case, skipping the permitting process entirely.

Watch the EPA's proposed rule change closely. If it goes through, the public's ability to even learn about a nearby facility before construction starts could disappear, which removes one of the few checks currently slowing projects down. Watch the state-level moratorium trend too. New York's move as the first state-level ban could either stay an outlier or become a template other legislatures copy as midterm pressure builds.

And watch the money. When a chipmaker is putting up $105 billion to guarantee a single data center lease, that's not a side bet, it's a sign of how central power access has become to the entire AI compute race. The GPUs are only half the equation now. The other half is whether you can find, permit, and power the building they sit in.