Let’s Talk About xAI and Data Center Pollution

So, some people are mad at me. That’s okay – I’m going to continue to call things as I see them. If you’re actually interested in the air quality issues in South Memphis – buckle up.

You might have seen my post from last night where I wrote about the xAI rally, and how it did not hit the mark in terms of coming up with any solutions. If not, you can find it on my profile.

There were lots of comments – too many to go through and answer individually – but the big one was – okay – what’s your solution?

That’s fair. And before I answer it, I want to make clear that I don’t doubt for one second that people are suffering. What I can’t tell you – what nobody can tell you right now – is whether that suffering is caused by xAI, by the refinery, by the steel mill, by the coal ash, or by the ozone this whole region has failed on for years. And here’s a hard truth about how human beings work: when you learn that machinery moved in next door, every cough starts to feel like evidence. That’s not a character flaw. That’s normal. It’s also exactly why we measure things instead of guessing.

South Memphis has earned every ounce of its distrust the hard way – a pipeline company once called this community the path of least resistance, the old Defense Depot leaked poison for decades, and promises made to these neighborhoods have been broken for longer than I’ve been alive.

So people deserve better than my comfort or my assumptions.

They deserve the truth about what’s actually in their air and where it’s coming from, because that’s the only version of caring that leads anywhere. If it’s xAI, we need to be able to prove it.

This community has been fed bad information by companies, by agencies, and by politicians for fifty years. The most respectful thing I can do is not add to the pile. So I spent this week inside the documents, and I want to show you what I found.

It started with a different comment. Somebody told me the EPA had tested the air here and found clear pollution problems. That surprised me, because I’d been following this fight for a while and I’d never seen an EPA test. So I went looking for it.

I found no public record of one.

As far as the public record shows, the EPA has never done air testing around the xAI site. What I found instead were the four things people keep mistaking for it.

There’s the EPA’s emissions inventory – a database where factories report their own pollution numbers. An analysis of the 2017 version found Southwest Memphis facilities accounted for 94% of the county’s reported factory emissions of the six main tracked pollutants [1].

That’s a factory-reporting database, not air testing, and it’s from 2017, six years before xAI existed. There’s an EPA rule from Jan 2026 about gas turbines – the machines that power the place, basically jet engines that burn natural gas to make electricity. The rule clarified that turbines mounted on trailers can still count as regular pollution sources that federal standards apply to. It was a rule change, not a test, and the EPA said it doesn’t apply backward in time [2].

There’s the American Lung Association giving Shelby County an F for ozone – the main ingredient in smog – and calling it the worst county in Tennessee for it. But that grade is built from existing monitor data, not new testing [3].

And there’s a community sensor project that gets called “the EPA study” online, except it was run by MCAP and a research group called CEEJH, not the EPA [4].

I know how it lands when somebody corrects a claim like that. It can sound like I’m defending xAI. I’m not. I’m looking for solid data that is actionable.

So, I decided to read every air study that actually exists for this neighborhood. There are four. It didn’t take long, because that’s how thin the record is for the loudest environmental fight in Tennessee.

The city’s test came first, it was done in June 2025. Memphis hired an outside lab called EnSafe to sample the air at three spots, including Boxtown, for about 12 hours on each of two days. They checked ten pollutants and reported nothing above the safety levels they compared against [5].

Then I read the criticism, and most of it holds up. Two days is a snapshot, not a picture of what people breathe year round. The test skipped ozone completely. The wind wasn’t blowing from xAI toward the Boxtown sampler on either day, and that sampler sat tucked near a building in a way that may not have followed the EPA’s placement guidance [6][7].

There’s one more problem that got less attention – some of the comparison levels the city used were workplace safety limits, the kind written for healthy adults on a shift, not for babies and grandparents breathing the air around the clock. The Southern Environmental Law Center put out a statement calling the results irresponsible [8]. On the setup, they had a point. If that test was meant to reassure people, I understand why it did the opposite.

Then I read the U of M analysis, and this is the one that deserves more attention than it gets.

Professor Chunrong Jia’s team built a computer model of the turbines’ pollution, then checked it against satellite readings of fine particle pollution – specks of soot so tiny they slip past your nose and throat and lodge deep in your lungs. They compared readings from before and after the data center opened. Their finding, published in The Conversation: the turbines “had not measurably degraded air quality in the surrounding neighborhoods,” with the model estimating xAI’s normal operations add about 1% to the particle level [9].

Two honest caveats, both flagged by the researchers themselves. The model ran on pollution numbers xAI reported about itself, which Jia admitted carry huge uncertainty, and it assumed the turbines ran normally the whole time, with no breakdowns. And Jia was blunt about the part that should reframe this whole fight – particle pollution in that area was already over the federal limit before xAI showed up. It still is. The people there weren’t breathing clean air that a data center ruined. They were breathing dirty air that nobody in power treated as an emergency until an Elon Musk came to town.

The TIME analysis came next, and it took me a minute to understand why both sides quote it. TIME magazine paid University of Tennessee researchers to measure a gas called nitrogen dioxide using satellite data. Nitrogen dioxide comes off anything that burns fuel – car engines, power plants, and these turbines – and it’s one of the main ingredients that cooks into smog.

The researchers measured two different things, and this is the part that matters. First, the average – the normal level of the gas over time. Second, the peaks – the worst moments, what are called the spikes.

The average barely moved after the turbines started, up about 3% once they adjusted for weather. The peaks told a different story. Inside the Boxtown neighborhood, the worst spikes ran 9% higher. In the satellite grid squares immediately around the data center, the worst spikes ran 79% higher [10].

Put those numbers side by side and they make physical sense – pollution is strongest near the machines making it and thins out as it drifts. But here’s the limit the researchers themselves flagged: a satellite can see that the gas went up, but it can’t tell you whose smokestack it came from. TVA’s Allen power plant sits right net to xAI – I’ve driven over there and have seen how close they are to one another. It ramped up its own generation in the same window, and the road between xAI and TVA is full of 18 wheelers wizzing by that never stop.

So, all the numbers are real, they all sit in the same analysis, and none of them can be specifically attributed to one thing.

The last one was the community project, and it held the biggest surprise for me. MCAP and CEEJH ran low-cost air sensors around South Memphis for four winter months. Their headline finding: particle readings frequently above 9 micrograms per cubic meter, the level the federal yearly standard is set at.

How often depended on where you stood – about 44% of hours at their Boxtown sensor, 75% at Westwood/Whitehaven, 33% at their Southaven site, and 62% across the network as a whole [4].

One technical honesty note, because it matters: the federal particle standards are a yearly average and a daily average, so you can’t call any single bad hour a legal violation. What those readings show is a neighborhood breathing elevated particle levels a lot of the time, which is reason enough to demand real monitoring.

And buried in their own report was a detail almost nobody covered. Their winter ozone readings – the exact pollutant everybody hammered the city for skipping – came in generally lower, with periodic spikes, and the researchers themselves said summer data is needed before anyone can draw conclusions [4][11].

Their critics have fair points too. The sensors aren’t the kind the gov’t accepts for official readings, the health dept says that type can read high in cold damp weather – the researchers say they corrected for it – but four months of data in a neighborhood with a refinery, a steel mill, and a gas plant can’t tell you which smokestack the particles came from [11]. That last part matters more than people realize. If the particles are coming from the refinery, then shutting down xAI wouldn’t help the family breathing them one bit.

I also went looking for the health numbers, because surely somebody has tracked whether people actually got sicker after the turbines came on. I found no public study of it – no before-and-after look at ER visits, hospital stays, or school absences.

The frightening cancer figure everybody quotes traces back to a 2013 U of M study that measured toxic chemicals in Southwest Memphis air and calculated that a lifetime of breathing them would carry a cancer risk about 4 times the national average [12].

That’s a risk estimate computed from measured toxins – not a count of actual cancer cases – and the air samples behind it were collected around 2008 to 2010, long before Elon Musk ever heard of Boxtown. And when I read the study itself, I found a detail that hardly ever comes up in this fight.

The researchers traced where that risk actually came from and it was mobile sources – the trucks, trains, barges, and planes moving through that corridor – drove 57% of it. Industrial sources drove 14% [12].

The most famous statistic in this whole fight, the one everybody points at the factories, was actually mostly traffic by its own authors’ math. That doesn’t excuse the factories, and one of the study’s top three risk chemicals, formaldehyde, is exactly what the turbines emit today – which is a reason to measure it, not a reason to relax.

But it should stop everybody cold for a second. The number this community has been organizing around describes a different villain than almost anyone assumed. That doesn’t make any of it less urgent. It makes it more urgent, because it means the problem is bigger than the villain everybody’s pointing at.

So – there are four studies. Two lean one way, two lean the other, and here’s the honest summary of all of them together. No study has been able to isolate what xAI specifically added to this neighborhood’s air, because that would take regulatory-grade monitors sitting near the site, and those don’t exist.

The satellite work did find nitrogen dioxide rising after the turbines started – but it can’t separate xAI’s share from the power plant next door. The particle work found no significant change – but it leaned on xAI’s own numbers. And nothing, anywhere, establishes that the air is safe. Every path through the evidence dead-ends at the same missing thing.

Now here’s something that should actually piss people off.

Effective Aug 1, 2024, the EPA awarded the Shelby County Health Dept $411,000 to build a real-time air monitoring station in South Memphis [13]. The concrete pad got poured Aug 14, 2025 – more than a year later. In March 2026, the health dept was running the legally required public comment process just to add the site to its official monitoring plan, aiming for May or June [14].

And here’s the detail that deflated me when I read the plan: the station is designed to measure carbon monoxide and ozone. Not fine particles. The pollutant at the center of the community’s strongest data isn’t in the plan [14]. One more thing I learned from the county’s own meeting records – ozone monitoring here is seasonal. The county’s plan runs the ozone monitor March through October, so even the pollutant the new station will track goes dark from November through February [14]. But it’s still not functional.

So this morning I did the simplest thing I could think of. I pulled up AirNow, the gov’t’s own air quality website, and looked at where Memphis’s official numbers were coming from. The fine-particle reading displayed for the Memphis area came from the county’s monitor at Shelby Farms. The ozone number came from Frayser, about 15 miles from Boxtown.

For more than ten years, there has been no official air monitor anywhere in South Memphis, and I found no public confirmation that the new station is reporting data yet.

That’s why every study is weak. That’s why both sides can wave data at each other forever. You can’t settle an argument about air nobody’s measuring, and the people caught in the middle of that argument are the ones breathing it.

One thing I should untangle before I get to the plan, because two different stories keep getting merged into one.

Colossus 1 sits in southwest Memphis near Boxtown. Its permit, issued July 2, 2025, covers 15 turbines with pollution controls [16]. You may have heard those turbines can only run 110 hours a year. But the 110-hour limit only counts startup and shutdown time – the dirtiest minutes, when the permit’s own emission limits don’t apply.

The permit contains no cap at all on total running hours. What limits them instead is a yearly ceiling on pollution emitted – no more than about 87 tons of smog-forming nitrogen oxides, for example [16]. You’ve probably also heard them called “backup” turbines. The permit never uses that word. Backup is how the company and the Chamber describe them, so I put in a records request so we find out what they actually do.

Colossus 2 is a different animal entirely. The building sits in the Whitehaven area of Memphis, but its power comes from across the state line in Southaven, Mississippi – and there are two separate turbine stories there that keep getting mashed together.

As of mid-July, Reuters counted 59 gas turbines running for Colossus 2 without federal clean air permits – at least 57 of them across the state line in Southaven – and xAI and Mississippi regulators say the portable units didn’t need them [17].

Separately, Mississippi approved a permit in March for 41 permanent turbines at the site – a future power plant that hasn’t been built yet [18]. The NAACP is suing over the unpermitted ones in a Mississippi federal court, and no judge has ruled on the merits yet [17]. Tennessee can’t issue or pull a Mississippi air permit, so that fight runs through Mississippi and the federal court. It’s a whole other can of worms, and everything in this essay is about the Memphis side.

I’ve made a list of who actually holds power over any of this, because the answer isn’t rally speakers, and it isn’t me unless you send me to Congress.

The Shelby County Health Dept issued the Colossus 1 permit and runs the monitoring network. The local Air Pollution Control Board handles appeals and variances when somebody challenges a permit.

The Memphis City Council passed the ordinance steering 25% of xAI’s city property taxes to the surrounding neighborhoods, and it could pass more [19].

The EPA enforces the Clean Air Act – the federal law that sets the limits on air pollution – and funds monitoring, usually through local health departments like ours. TDEC, the state’s environment dept, takes state-level complaints. Mississippi’s environment dept permits the Southaven turbines.

Federal courts are hearing the NAACP case. MLGW proposes the water and power rates, and the City Council approves them. The TVA – the gov’t-owned power company that supplies this whole region, with a board nominated by the President and confirmed by the Senate – plans how much grid power gets built here. And Congress oversees the TVA and the EPA and writes the laws for the next data center.

So what would I do about all of this? Once I could see the whole board, the plan wrote itself.

I already started on the first part. This week I filed two public records requests – formal demands that force the gov’t to hand over its documents. Any citizen can file one, and I’ll publish whatever comes back. One went to the health dept for records showing each turbine’s actual dates and hours of operation – the permit requires xAI to keep monthly fuel and pollution records, logs of every startup, shutdown, and malfunction, and to file reports twice a year [16] – plus every air complaint filed from the 38109 zip code.

That will replace dueling stories for the operating record. The other went to TDEC, because I’ve seen it reported that families near the site filed complaints with the state and got turned away. I can’t verify that yet, but it’s exactly what the records request is for – if families asked their gov’t for help and got told no, I want the record of it, and I want to know whether the reason was money, staffing, or something worse.

The second part is already in motion, and almost nobody noticed.

The construction permit told xAI to apply for a permanent operating permit once the turbines started up, and the company did – for 12 turbines instead of 15. The county posted that request back on April 7, in the legal notices section of the paper, with a 30-day public comment window and a plain statement that the health dept “intends to issue” the permit [20].

One detail in that notice ties this whole story together: the new permit would write the EPA’s Jan 2026 turbine rule directly into xAI’s operating requirements.

Here’s the part that should bother you.

That comment window opened and closed in the spring while everyone was arguing on Facebook, and more than two months later, no decision has been publicly reported. The quiet conversion of a temporary permit into a permanent one is the real next decision on the xAI that already exists – not any bill in Congress, and not the construction permit’s Jan 2, 2027 expiration date – and it’s moving through the system with almost nobody watching.

So I’m requesting the full file: the application, every comment that came in, the county’s answers, and the decision when it lands. The permit rules allow an appeal within 30 days of a decision, and when that clock starts, I’ll publish the deadline and show District 5 exactly how to be heard. If you want the permanent permit to require public disclosure of when those turbines actually run – instead of self-reporting twice a year – the appeal window is the next door in, and I’ll make sure you know the moment it opens.

The third part is the boring fix that settles everything downstream – a real monitoring network.

I mean the official kind: federal reference monitors run under an EPA-approved quality plan, placed upwind and downwind of the industrial corridor, measuring the pollutants actually in dispute – fine particles, nitrogen dioxide, ozone, and formaldehyde – with the data live on the internet. The station currently planned measures carbon monoxide and ozone, which means the fine-particle question stays unanswered even after it turns on [14].

This would protect South Memphis from every polluter around it, not just the famous one, and it’s the one thing every side of this fight says it wants. That makes it the easiest federal ask on the board, and I’d make the ask.

Fourth, I’d vote for the AI Data Center Site Selection Transparency Act, the bill introduced in the House this spring [21].

Here’s what it actually does – no more, no less. It makes developers disclose a proposed data center’s location at least 180 days before taking a definitive step toward building, disclose an estimate of the resources it’ll take to build it, provide an environmental impact analysis prepared by an independent third party the developer pays for, do real multilingual outreach in the affected community, and it bans the hush agreements between developers and gov’t bodies that kept the Memphis deal secret until the machines showed up.

It gives communities notice and a voice, not a veto – I want to be honest about that. But Memphis found out about the turbines via the news, and this bill makes that impossible next time. I’ve also said that if the AI Data Center Moratorium Act – the federal ban on new data center construction introduced by Rep. André Carson, with Sen. Sanders and Rep. Ocasio-Cortez behind it – came to a floor vote, I’d vote for it [22].

But I won’t pretend it does something it doesn’t. A ban on new data centers doesn’t remove one turbine, one permit, or one gallon of anything from the data centers here that are already built and running. The transparency bill covers the future. The operating permit, the records, and the monitors deal with the present, and the present is where those families live.

And the fifth part came straight out of that timeline. I’d make the agencies show they can actually do their jobs.

Congress should ask why an EPA monitoring grant awarded in Aug 2024 still hasn’t produced a working station, and whether the state agencies handling complaints have the staff and money to handle them – and if not, whether the gap is state money, federal money, or both.

The county has already told us this is fragile. At that same Sept 2025 meeting, according to a summary of it, the health dept’s deputy director said the monitoring program had been approved to continue into the next federal fiscal year, with the long-term effects of EPA policy changes still unknown [13]. Translation: the one station South Memphis is finally getting could lose its funding before it matters. Guarding that money is real congressional work. It doesn’t make headlines, and it’s exactly what a House member from this district should be doing.

Look, I know how I’m supposed to talk about this. I’m supposed to pick a side and crank the volume. But I keep thinking about something most of us claimed to believe until about five minutes ago – that a room full of people who all think the same makes worse decisions, and that different perspectives make us smarter.

A lot of folks defended that idea loudly when it had a name and a budget. Some of those same folks now treat every unapproved question as betrayal. I look at things differently than most people, and I’d argue that’s exactly what you want in a representative.

I know some people will read all this and decide I don’t care enough, because I didn’t call for a complete shutdown. But the truth is – there’s been fifty years of strong feelings from every direction, and Boxtown still doesn’t have one official air monitor. I’ve knocked doors there. People talked to me about getting to work without a car and the empty houses rotting on their streets, and they talked about it like people who’d stopped expecting anyone to follow through. The greatest disrespect is using their struggles as a platform rather than looking for the hard, unsexy solution.

So – to sum all this mess up – the air in South Memphis was bad before xAI came. It’s bad now. If xAI shut down tomorrow, it would still be bad, because the refinery, the steel operations, the coal ash left over from decades of burning coal, and more than 17 nearby facilities that report toxic releases to the EPA would still be there – and there still wouldn’t be an official particle monitor in South Memphis measuring any of it [23].

xAI ran turbines before it had its Memphis permit – that part is simple fact – and whether that broke the Clean Air Act is now for a court to decide, with real consequences if it did [17].

And the families near that fence line deserve better than being anybody’s backdrop. They deserve monitors that exist, records they can read, a permit process they can walk into, and a representative who tells them the truth even when a slogan would poll better. That’s what caring looks like when it has to survive past the news cycle.

Every source I used is listed below. In the end, my main goal is to find a real solution – so tell me what I got wrong

SOURCES

If you like facts-first info on all this, consider leaving a tip on Venmo or PayPal – this stuff takes a lot of time to put together. And check out everything I’ve written with citations and receipts at the xAI Files.

Sources

Every claim in this essay is numbered. The key documents are hosted here, not just on portals that can move them.

1. WHEJAC public meeting summary, April 28, 2021, pp. 59-60 – Justin Pearson quoting Sarah Macaraeg's analysis of the 2017 EPA National Emissions Inventory annotated excerpt the underlying database
2. EPA final rule, New Source Performance Standards review for stationary combustion turbines, Federal Register, Jan 15, 2026
3. American Lung Association, State of the Air 2026, Tennessee release
4. MCAP/CEEJH, follow-up report for the hyperlocal air quality monitoring network in South Memphis
5. City of Memphis, initial air quality testing results, June 24, 2025
6. TechCrunch, "Air quality tests around xAI's Memphis data center raise questions," June 25, 2025
7. MLK50, "Flawed air quality test leaves community twisting in the wind," July 3, 2025
8. FOX13 Memphis, "City releases air quality test results; xAI, opponents respond," June 25, 2025 (the SELC statement)
9. Jia, Titu & Batbaatar (University of Memphis), The Conversation, Sept 23, 2025
10. TIME, "Inside Memphis' Battle Against Elon Musk's xAI Data Center," Aug 2025 (UT-Knoxville satellite analysis)
11. Tennessee Lookout, "Shelby Co. Health Department questions integrity of community air monitoring report data," June 23, 2026
12. Jia & Foran, "Air toxics concentrations, source identification, and health risks: An air pollution hot spot in southwest Memphis, TN," Atmospheric Environment (2013)
13. Shelby County Board of Commissioners, Hospitals & Health Committee, Sept 3, 2025 – SCHD monthly update on the EPA-funded monitoring station
14. SCHD, modified 2025 Ambient Air Monitoring Network Plan (Riverview site; March 17, 2026 public notice) notices page
15. AirNow.gov, Memphis reporting area, checked July 18, 2026 screenshot: the monitor map airnow.gov
16. SCHD Air Permit 01156-01PC, July 2, 2025 annotated copy
17. Reuters, "Pollution from Musk's unpermitted xAI power project hits hardest in Black communities," by Disha Raychaudhuri and Valerie Volcovici, July 14, 2026 the complaint (PDF)
18. Mississippi DEQ Permit Board approval of 41 permanent turbines, March 10, 2026, as reported
19. Memphis City Council Ordinance 5953 (25% of xAI city property tax to surrounding communities)
20. SCHD public notice, April 7, 2026 (notice Md113657) – the operating-permit conversion request – original legal notice notice excerpt (PDF)
21. H.R. 8488, AI Data Center Site Selection Transparency Act of 2026
22. Rep. André Carson, AI Data Center Moratorium Act announcement, June 2026
23. E&E News by POLITICO, Southwest Memphis TRI facility count
24. Mississippi Dept of Environmental Quality, complete public records production on air permit 0680-00119 (MZX Tech, LLC, 2875 Stanton Road, Southaven), received July 28, 2026 – portable turbine inventory filed by Trinity Consultants June 12, 2026; start-of-construction notices April through July 2026; MDEQ correspondence May 6, 2026. Obtained by public records request; documents on file.

Correction, September 27th, 2026: This essay first said the 2017 emissions data was seven years before xAI existed – it was six. It said the ozone monitors shut down from October through March – the county’s plan runs them March through October. It put the words “can only run 110 hours a year” in the permit – the permit caps startup and shutdown time at 110 hours a year. It said the House bill requires estimates of power and water – the bill says the resources needed to build. It put the words “remain to be seen” in quote marks as the county deputy director’s – they come from an AI-written summary of the meeting. And it said the fine-particle reading came from a monitor in Marion, Arkansas – the AirNow screenshot shows the county’s Shelby Farms monitor.

Update, September 27th, 2026: A few lines now say who says so, and “air monitor” now says “official” where community sensors exist. The county’s new South Memphis station started posting ozone readings on August 16th, 2026, after this essay ran.

Everything I have written about AI is in one place at rachelandthecity.com/the-ai-files.