The Data Center Bill

As the midterms approach, black-and-white politics is in full swing. Someone is evil — take your pick: billionaires, immigrants, AI, MAGA — and they're all coming for the good old average American, the middle class. (Never define middle class. That way everyone finds themselves in it.)

One of the hot topics lately: data centers. And it's bipartisan. In this climate that should feel like a win — one thing our politicians and most Americans actually agree on.

And yet.

I had a feeling data centers could be a good thing for the economy and for communities. So I set out to research it, planning to write an article about what I found. The research turned out good enough on its own, so here it is — the state of the argument, as far as I can establish it, with everything sourced at the bottom.

It's long. I think it's worth it.

1. The political moment

The short version: this became a bipartisan issue, which almost nothing is, and it happened fast.

Gallup found roughly seven in ten Americans oppose a data center being built near them. A late-July Fox News poll found the same 70 percent — and 78 percent support slowing construction to address environmental and community concerns.

The opposition is not partisan. 60 percent of Republicans, including 53 percent of MAGA Republicans, oppose local construction. In the first quarter of 2026 alone, local opposition blocked or delayed 75 projects worth $130 billion, according to Data Center Watch — as many as in all twelve months of 2025.

The National Republican Senatorial Committee has flagged data centers as a sleeper issue for the entire midterm cycle. Candidates in Pennsylvania, Michigan and Ohio are distancing themselves from projects they previously courted. And more than 300 state bills were filed across 30-plus states in the first six weeks of 2026 — with the character changed: from incentives to attract data centers, to regulation to control them.

The era of "data center equals economic development win" ended somewhere around 2025. Not everyone building has priced that in yet.

2. What's actually driving bills up

This is the contested part, and most coverage picks a side.

Both are true, and they are answering different questions. Nationally, data centers are one driver among several. In PJM specifically — the grid serving some 65 million people across the mid-Atlantic and Midwest — they are the dominant marginal driver.

The politics are hottest exactly where the localized math is worst: Virginia, Ohio, Pennsylvania, Georgia. That geographic concentration is the story, and it's the part most takes miss.

3. The solution menu

Six families of solution. Most are further along than the news implies.

Large-load tariffs — the main event

A separate utility rate class for very large customers, designed so they pay the true cost of serving them rather than socializing it.

23 states have approved at least one large-load tariff as of May 2026, with seven more pending, according to the Edison Electric Institute. At least 18 states have introduced legislation creating special rate classes.

The mechanics are more serious than "pay your share":

• Minimum take-or-pay. AEP Ohio requires customers above 25 MW to pay for at least 85 percent of subscribed capacity for up to 12 years, regardless of actual use. That kills the reserve-500-use-100 problem.
• Long contract terms. Virginia's GS-5 tariff requires 14-year contracts above 25 MW, plus $1.5 million per MW in collateral. Georgia extended allowable terms from 5 to 15 years for loads above 100 MW.
• Full upgrade costs. The customer covers the distribution, transmission and generation upgrades their project triggers.
• Exit fees and collateral, so a bankrupt or relocating tenant doesn't strand the asset.

Recent adopters include the Oregon PUC, the Pennsylvania PUC in April 2026, and Ohio — where the tariff was upheld after data center firms tried to overturn it.

Flexible load and curtailment — the most underrated

The most interesting finding in the whole file, and one that barely surfaces outside energy circles:

Curtailing just 0.25 to 1 percent of annual data center consumption during critical hours would let the existing grid absorb up to 100 GW of new load without major buildout. Average curtailment event: about two hours. Duke University · Nicholas Institute for Energy, Environment and Sustainability

That reframes the problem. The constraint isn't total energy — it's peak coincidence. And AI training workloads are unusually shiftable compared to, say, a hospital.

It isn't theoretical. EPRI's DCFlex field tests are running now: Google and Duke Energy in North Carolina on workload scheduling, and Oracle and NVIDIA in Arizona achieving a 25 percent AI cluster power reduction sustained for three hours, with no hardware retrofits. A separate report puts unlockable headroom at 76 GW through optional curtailment.

The catch: data centers are contractually and culturally built around five-nines uptime. Getting them to accept interruptibility is a commercial problem, not a technical one.

Bring-your-own-power

Make the data center build its own generation instead of drawing from the shared grid. 56 percent of developers are exploring co-located or on-site generation, per the Foley 2026 survey, and around 30 percent of sites may use on-site power as their primary source by 2030. FERC ordered PJM in December 2025 to establish formal rules for direct connection to generation.

The part that gets skipped: if it's a gas plant, it bids up regional gas prices — which raises the marginal cost for every gas generator on the system. That pass-through happens entirely outside any tariff structure. Virginia is also litigating whether behind-the-meter turbines let data centers skirt state clean energy mandates. Bring-your-own-power looks like paying your own way, and often isn't.

Community benefit agreements

Legally binding local deals: water replenishment, local hiring, workforce training, noise mitigation, funding for schools and infrastructure in exchange for tax abatement.

The live fight is transparency. Most negotiations happen under NDA between developer and county officials, so residents can't influence terms before they're locked. Good Jobs First argues the core problem is decades-long property tax abatements — companies this profitable should pay full freight, and where abated, should fund public goods directly.

Moratoria and rolling back tax incentives

Moratorium bills have been introduced in New York, South Dakota and Oklahoma to pause construction pending impact studies. Virginia, Georgia and Oklahoma are proposing to cut or eliminate the tax credits that attracted these facilities in the first place — mostly the same states that competed hardest to win them.

Federal proposals

The SHIELD Act would update federal utility policy so massive users, not ratepayers, bear grid infrastructure costs. The Ratepayer Protection Act would require state regulators to set a large-load standard for users at or above 100 MW, directing them to cover the full incremental cost of generation, transmission and local infrastructure.

Neither is likely to move before the midterms. State public utility commissions are where the action is.

4. Where the fixes fall short

The part that keeps this from being a happy story. Consumer advocates' objections to large-load tariffs:

1. The proof problem. If you can't prove a cost belongs to the large load, it defaults into the shared revenue requirement — meaning residential customers absorb it. Cost allocation is a documentation problem, and the documentation regimes are weak.
2. Future-growth carve-outs. Utilities don't want upgrades serving future load counted against the current large-load customer. So residents get billed for infrastructure built on a forecast, and if the forecast is wrong, they're stuck with it.
3. Speculative forecasts. The killer example: AEP Ohio cut its own large-load forecast from 30 GW to 13 GW — more than half — after its tariff was approved. Utilities are proposing capital investment against numbers one analysis called pure speculation.
4. Voltage loopholes. Future industrial customers may be able to escape the large-load class simply by taking service at high voltage.
5. Tariffs don't touch fuel costs. No tariff structure addresses gas price pass-through from new gas demand, which hits every ratepayer regardless.

Approving the tariff was the easy part. Whether it works is an evidentiary and enforcement question most states haven't solved.

Where that leaves me

I went looking for whether anyone had proposed a way to build this infrastructure without landing the cost on people who never asked for it. The answer is yes, in more detail and further along than the coverage suggests — 23 states, real contract terms, real collateral requirements, and a curtailment finding that would change the arithmetic if the industry accepted it.

What I can't tell you is whether it works, because the evidence for that doesn't exist yet. The tariffs are new, the enforcement is untested, and the load forecasts they rest on have already been revised by half in at least one state. That's not a verdict either way. It's the honest position, and it seemed worth writing down while everyone else is picking a side.