The question nobody can dodge anymore
When we first published this piece in September, the question sounded almost academic. Who should pay for the power plants, power lines and substations that AI data centers need?
Six weeks later, it is the most practical question in American energy. Voters in several states just weighed in at the ballot box. Utility commissions are writing new rate classes. Congress tried to settle it and came up three votes short in the Senate.
The core tension hasn't changed. A single hyperscale campus can need 300 megawatts to more than a gigawatt. That's the load of a mid-sized city, often arriving in two or three years. The grid wasn't built for that pace. Someone has to pay for the generation, transmission and distribution upgrades that make it possible.
The old answer, socializing the cost across every customer, is breaking down. Here's why, and what is replacing it.
Why this fight got so loud
Utilities have historically treated new load as good news. More customers spread fixed costs over more kilowatt-hours, which can push rates down for everyone. That logic works when new load is predictable and spread out.
AI load is neither. It's concentrated, it's huge, and it's uncertain. Developers file speculative requests in several places for the same project. In PJM, the grid operator recently trimmed its large-load forecasts by about 6.5 gigawatts after scrubbing duplicate and doubtful requests. If a utility builds a $1 billion substation and transmission package for a campus that never shows up, or leaves after five years, the leftover cost lands on everyone else.
Households are already feeling price pressure. EIA's October 2026 Short-Term Energy Outlook shows wholesale power prices up about 11% nationally and about 41% in PJM from a year earlier. Not all of that is data centers. Fuel costs, aging infrastructure and capacity shortfalls all play a part. But the timing has made data centers the face of the problem. That's a big part of the public trust problem we've written about before.
Congress tried, and stalled
In September, the House passed H.R. 9340 by a 417-3 vote. That's a rare level of agreement in this Congress. The bill amends PURPA, the 1978 law that shapes how states regulate utilities. It requires states to consider whether customers with 100 MW or more of load should pay the full incremental cost of the generation, transmission and distribution they cause.
Senator Jon Husted of Ohio summed up the idea: "if you create the cost, you should pay the cost."
The Senate path was rougher. On September 17, a unanimous consent request to pass the bill quickly was blocked. Senator Martin Heinrich called for "real legislation with real teeth" and pushed his own GRID Savings Act. In late September, a cloture vote failed 57-43, three votes short of the 60 needed. Senator Catherine Cortez Masto argued the bill was voluntary and unenforceable, because it only requires states to consider the standard, not adopt it.
That criticism is fair. Even if it becomes law, H.R. 9340 doesn't force any state to change a single rate. It sends a strong signal, but the actual decisions still happen in state commissions and at FERC.
So for now there is no federal rule. The decisions are being made elsewhere, and quickly.
FERC is setting the ground rules
On June 18, 2026, FERC issued show-cause orders to six regional grid operators. FERC regulates wholesale markets and interstate transmission. The orders asked each operator to explain or fix how it handles large loads in five areas:
- 1.Interconnection study processes, and why they take so long
- 2.Preventing cost shifting to other customers, with more transparency
- 3.Colocation, where a data center sits right next to a power plant, and behind-the-meter arrangements
- 4.Transmission service for flexible loads that can reduce demand on request
- 5.How colocated generation gets studied
In a September 29 concurrence, Commissioner David Rosner went further. He urged states to adopt their own large-load tariffs. He also backed zonal cost assignment in PJM, so that costs land closer to the regions where data centers actually connect, not on the whole 13-state footprint.
FERC can't set retail electric rates. That's the states' job. But it can decide who pays for transmission and how wholesale capacity costs get allocated. Those are two of the biggest line items behind your bill.
The states are writing the real rules
This is where the action is. According to DSIRE’s DELTa tracker, large-load tariffs and service rules jumped from 41 in July 2025 to 104 in July 2026. Of those, 69 are approved and 35 are proposed, across 37 states and more than 70 utilities. About 45% use a threshold of 50 MW or more. Roughly a quarter include some kind of flexibility option.
A few examples show the range:
- California. On September 22, Governor Newsom signed laws that create a separate rate class for large data centers and require upfront payment for grid upgrades they trigger.
- Virginia. On September 18, Governor Spanberger laid out a framework in the world’s largest data center market: an executive order now, with legislation planned for 2027.
- Texas. SB6 requires large loads to share interconnection costs and to be curtailable during grid emergencies. ERCOT and the PUCT are turning that into detailed rules.
- Ohio. AEP Ohio’s data center tariff makes new large customers pay for most of their contracted demand even if they use less. We covered that in depth in Ohio.
- North Carolina. Duke Energy’s large-load proposal is before regulators, with a decision expected in mid-November. See our North Carolina breakdown.
- Pennsylvania. The state has published a model tariff that utilities can adapt.
- Minnesota. Xcel offers an interruptible rate that rewards data centers for cutting load when the grid is stressed.
The new playbook: “you cause it, you pay for it”
Across all these tariffs, the same tools keep showing up:
- Minimum demand charges. The data center pays for a large share of its contracted capacity, often 60% to 90%, whether it uses it or not.
- Long contract terms. Commitments of 10 to 20 years, so a utility isn’t left holding stranded assets.
- Exit fees. If a project leaves early, it pays the remaining cost of the infrastructure built for it.
- Upfront payments or collateral. Cash, letters of credit or guarantees before construction starts, which also weeds out speculative requests.
- Separate rate classes. Data center costs and revenues are tracked separately, so subsidies are easier to spot.
The goal is simple. If the load shows up, it pays its way. If it doesn’t, the developer, not the household, eats the cost.
Flexibility is the new bargaining chip
The most interesting shift is how flexibility is changing the deal. A Duke University study estimated that the U.S. grid could absorb roughly 76 GW to 126 GW of new large load if those loads curtail a small share of their annual hours, between 0.25% and 1%. That means using existing headroom instead of building new peak capacity first.
That turns a data center from a pure cost into a potential grid asset. Tariffs are starting to reward it with faster interconnection, lower demand charges or interruptible rates. On-site batteries, backup generation and shiftable training jobs all help. It’s the same playbook Bitcoin miners have used for years, which we covered in Bitcoin miners and the 2027 deadline.
It also changes the economics of the AI factories that will need hundreds of megawatts each.
Who is deciding who pays
| Player | What they did | Status | What it means for your bill |
|---|---|---|---|
| Congress | H.R. 9340: states must consider charging 100 MW+ loads the full cost of new generation, transmission and distribution | Passed the House 417-3. Senate cloture failed 57-43. | No federal guarantee yet |
| FERC | Show-cause orders to six RTOs on studies, cost shifting, colocation and flexible load | Open, with responses under review | Sets the wholesale and transmission ground rules |
| States | Separate rate classes, upfront upgrade payments and model tariffs (CA, VA, TX, PA, OH, NC) | 104 tariffs and rules, 69 approved | The biggest real protection today |
| Utilities | Minimum bills, 10–20 year contracts, exit fees and interruptible rates | Spreading across 70+ utilities | Shifts risk from households to the data center |
| Tech companies | Voluntary pledges and community compacts | Voluntary and unenforced | Helpful, but not a substitute for tariffs |
What companies are promising
The tech industry sees the backlash. In July, major companies signed a voluntary Ratepayer Protection Pledge. OpenAI has since rolled out its National Community Compact, committing to cover its own energy costs in host communities.
These are good signals, but they are promises, not rules. A tariff is enforceable by a utility commission. A pledge isn’t. That gap is why the states keep legislating even as companies volunteer.
So who should pay?
Our view hasn’t changed since September, and the evidence has gotten stronger.
Data centers should pay the full incremental cost of the grid upgrades they cause. That includes generation, transmission and distribution, backed by contracts long enough to protect everyone else.
Shared benefits should be shared fairly. If a new transmission line also improves reliability for a whole region, it’s reasonable to spread part of that cost. But that has to be shown in the record, not assumed.
Flexibility should be rewarded. A data center that can drop load at peak is cheaper to serve and should be priced that way.
Transparency should be required. Ratepayers should be able to see what data centers pay and what the grid spends on them.
The AI Power Wall won’t be solved by one law. It is being solved one tariff at a time. If that work is done well, the AI boom can pay its own way and possibly lower costs for everyone else. If it’s done badly, households will subsidize the biggest companies on earth, and public support will fall apart.
Chris Kalowes
Founder, WattThe?!
Energy intelligence. Simplified.
