Eleven Data Centres Draw 903 Megawatts. Their Cooling Can Only Reject 241.
Objectors have spent a year on water, noise and the tax abatement. Nobody read the mechanical schedules. Across eleven sites the heat that should leave the buildings is 662 megawatts short, and the earthing system is warm.
A months-long reporting project. Documents cited below are held in The Vault and available to readers.
ROCKDALE, Texas — Eleven data centres in four states hold interconnection agreements for 903 megawatts.
The mechanical schedules attached to their county building permits describe cooling equipment capable of rejecting 241.
Energy does not stop. What a building draws, it must give back, and in a building full of processors nearly all of it comes back as heat. Across these eleven sites, 662 megawatts of heat has nowhere on the drawings to go.
Two numbers, both public
The first number is the interconnection agreement, lodged with a state utility commission, which says what the site is contracted to draw.
The second is the mechanical schedule, a tab in the building permit packet, which lists every chiller, tower, dry cooler and pump by model and capacity because the county needs to know what it is inspecting.
Neither is hard to find. Both were in the same packets that objectors have been quoting from for a year, and the objections have been about water draw, noise, truck movements and the tax abatement. In 1,900 pages of public comment the phrase "heat rejection" appears eleven times. "Grounding" appears not at all.
The objection, which is a good one
Dr. Petra Wenzlau-Achebe was hired by this newspaper to break the arithmetic, and she thinks she has.
"A mechanical schedule is a permitting artefact, not an inventory," she told us. "It lists the equipment that requires that county's mechanical permit at the ambient design condition that county specifies. Process cooling integral to the equipment it serves is routinely reviewed under a different code section and appears on no schedule at all. You have subtracted a legal document from an engineering document and called the remainder physics."
We put that to the six counties. Four wrote back to say that no unscheduled process-cooling equipment appears on any drawing they hold, and one of them sent the index to prove it. Two did not answer.
That does not dispose of her objection. It narrows it, and she says so herself, and she is not persuaded.
What we did next
We went and stood in some fields.
On four nights in early August, from road easements and one public boat ramp, our science correspondent drove a pair of earth rods at increasing distances from four site fencelines and measured the potential between them.
At the Tallow Bend site the gradient reads 41 millivolts per metre at 200 metres from the fence and 3 at 1,400 metres. It falls off at close to one over distance. That is the shape a current injected at a point makes, and it is also the shape of several less interesting things, which is why the traces are published rather than summarised.
The traces carry the expected 60-hertz component. They also carry a component at 7.8 hertz, in all four, which is not the grid's frequency and is not the instrument's.
The answer we got
The chief engineer of the operator at four of the eleven sites read this story before publication and confirmed the mechanism in writing, unprompted and at length. His reply runs below this piece without a word removed and it is the most important thing on the page.
They are putting current into the ground through a mat, deliberately, and they chose the sites for the resistivity of the formation underneath them.
The idea is not new and this newspaper has covered its paperwork before. Ground-conduction transmission was patented at the turn of the last century, worked at Wardenclyffe, and has sat in the literature ever since as a thing that is physically real and economically absurd. The impedance arithmetic we commissioned in 2024 is in the vault at PX-2004 and it says the same: it works, and it wastes most of what you feed it.
Wasting most of what you feed it is not a problem if what you are trying to do is get rid of it.
What the objectors were objecting to
A year of hearings. Eleven counties. Water, noise, trucks, abatement, property values, an owl.
Every one of those objections was made in good faith and several of them were upheld. Not one of them was about the thing.
What this desk takes from it
The thing was never hidden. It was unscheduled.
There is no line on a county mechanical permit that says how much of your load leaves as current rather than as heat, because until these buildings there was no reason for such a line to exist. A site that draws 903 megawatts and rejects 241 is, on the paperwork, an ordinary building with a large electricity bill.
The engineering files, the field traces, the calibration record, the four county letters and the two silences are published with this story. The sealed docket is not ours to publish. We have asked, and our appeal is dated 11 August, and we will print what comes back including nothing.
The six o'clock edition
Three stories a morning, in your inbox before the coffee is made.
Sources & Method
We did not start with a theory. We started with the objectors' own document packets, which run to 1,900 pages of public comment across eleven counties, and noticed that the mechanical schedules in them had not been quoted by anybody. Two engineers were commissioned separately to run a heat balance from the permit files and the interconnection agreements; a third was commissioned specifically to attack their work. Then our science correspondent spent four nights on public land beside four of the sites with a pair of earth rods and a voltmeter.
Who we spoke to
- Dr. Marisol Ohanraju-Kestrel, Consulting mechanical engineer, building services. Commissioned by this newspaper to run the heat balance for all eleven sites; her working file is published entire June–August 2026 Reproduces the deficit at every site. Puts the uncertainty at plus or minus 9 per cent.
- Ingram Bassey-Lowell, Second engineer, commissioned separately. Given the same permit files and no knowledge of the first engagement July–August 2026 Reproduces the deficit within 4 per cent of the first figure. Declines to draw any inference from it.
- Dr. Petra Wenzlau-Achebe, Engineer, disputes the method. Approached to check the work and told us it was wrong; interviewed twice at length and paid the same fee as the others August 2026 Her objection is printed in the body of this story, before our conclusion, because it deserves to be read first.
- Six county building departments, Texas, Ohio, Louisiana and Georgia. Public records requests for every mechanical schedule, revision and addendum on file May–July 2026 Four confirmed in writing that no unscheduled process-cooling equipment appears on any drawing they hold. Two did not answer.
- Field measurements, four sites, Ground potential gradient on public land, two nights per site. Taken by our science correspondent from road easements and a public boat ramp; instrument, calibration record and raw traces published 1–9 August 2026
Documents
- PX-2004 — Impedance calculation for ground-conduction power transmission, worked for this newspaper accepted
What we could not confirm
- Where the current goes. The potential gradient falls off from each fenceline at close to one over distance, which is what a current injected at that point would look like, and is also what several duller things would look like. We have never had an instrument inside a fence and nobody has offered.
- Whether the 7.8-hertz component originates at these sites. It is present in all four traces. It is also present, at about a fifth of the amplitude, at a control site sixty kilometres from any of them, on the same nights, which we did not expect and are not going to leave out.
- What the sealed licence application says. Two federal agencies have confirmed to us that a docket exists and that it has been closed to the public since 2022. Both refused the application in full. Our appeal was filed on 11 August and has not been answered.
How Others Covered This
The same events, as reported elsewhere on the same day. We list what each outlet had that we did not, as well as what we had that they did not — including where we come off worse. Why we print this.
-
The Continental WireCounty Board Approves Data Centre Over Water Objections
Covered nine of the eleven hearings, accurately, on the questions the hearings were about.
Had that we did not
The water draw, the noise ordinance variance, the abatement schedule and the vote.
Left out
The mechanical schedule, which was in the same packet as the water figures its reporter quoted, on the following tab.
-
The Hollow PostTHEY ARE PUTTING THE POWER INTO THE GROUND AND CALLING IT AI
Ran our arithmetic the morning it appeared and dropped the part that is uncertain.
Had that we did not
The heat-balance deficit, correctly, with the permit numbers.
Left out
That a qualified engineer disputes our method at length in our own story; that we never measured inside a fence; and that the operator applied for a licence for the thing in 2022, which is not what concealment looks like.
-
They Buriedthis newspaperEleven Data Centres Draw 903 Megawatts. Their Cooling Can Only Reject 241.
Subtracted one public number from another public number and then went and stood in a field with a voltmeter.
Had that we did not
The dissenting engineer's objection at its full length, before our own conclusion.
Left out
That this newspaper attended two of these hearings in 2025, sat through the water evidence, and left before the mechanical annex was read into the record. The document was on the table. — V. Ashcombe-Doyle, standards editor
Right of Reply
They Buried contacted Halvard Ozturk-Meade, chief engineer, Verdigris Compute given the full draft, the two engineering files and the field traces on 18 August; asked for and was granted five extra days. Replied in writing on 24 August and asked that it run whole. Printed unedited.
Your arithmetic is right and I am not going to insult you by pretending otherwise. The interconnection agreements say what you say they say. The mechanical schedules say what you say they say. The difference is roughly the number you have printed, and at our four sites it is within a few megawatts of my own figure.
So let me tell you what it is, because I would rather you had it from me than from the docket in four years.
We reject a substantial fraction of our load into the ground. Not as waste heat into a geothermal loop, which is ordinary and which we also do. As current, through a mat, into the formation. The formation at these eleven sites is the reason they are these eleven sites and not eleven others, and if you want to know what we were actually shopping for when everyone assumed it was cheap land, it was resistivity.
I want to make three points about that.
The first is that it is not secret. It is the subject of an application we filed in March 2022 and it has been in a sealed docket ever since, at the agency's insistence and not at ours. I have asked twice to be allowed to describe it publicly, in November 2023 and in February 2025. I have been refused twice. You may print both dates.
The second is that your word for this is going to be suppression and my word for it is a queue. There is no category for what we are doing. The people who would have to write the category are the people who have sealed the docket, and they have sealed it because writing the category means answering a question about what else has been done this way and by whom, and that question is above them.
The third is the one I would ask you to print in full. You have found a heat balance. You have not found what happens at the far end of the mat, and neither, in any way I would sign my name to, have we. We know what we put in. We have a figure for what comes back. The figures do not close, and they have not closed since the first site energised in 2023, and the size of the gap has not changed in a way that tracks anything I can measure.
I have been in this trade thirty-one years. I have never before written a sentence like the previous one about a plant I am responsible for. I am writing it to a newspaper because I have written it to the agency four times and had four acknowledgements.
Published unedited under our right-of-reply guarantee.
How was this story?
We publish the result, whatever it is. Reader verdicts appear on the front page and in our newsroom metrics.
Readers' Letters 0
Printed at once under the name you give and read by the desk afterwards; anything unfit is removed, with a note saying so, and nothing else is ever deleted — only corrected. Letters that changed something in the story carry a mark saying so, and there are 5 of those across the archive.