The Black Garden

The University of Chicago recently published one of the better papers I've read on using geoexchange borefields to cool AI data centers. I believe they're right.

As artificial intelligence pushes electrical demand into the gigawatt range, conventional cooling is becoming as much of a constraint as power itself. Underground Thermal Energy Storage (UTES) offers a compelling answer: use the earth as a seasonal thermal battery, storing cold when electricity is plentiful and recovering it when cooling demand peaks. The thermodynamics are sound. The potential is enormous.

Then I reached one number. To cool a single one-gigawatt data center, the study models approximately 686 vertical boreholes, each roughly 275 meters (900 feet) deep.

One building. Six hundred eighty-six wells. Now multiply that across the hyperscale buildout already underway. The future isn't hundreds of boreholes. It is hundreds of thousands.

And that is where my concerns begin — not with the thermal model, but with the construction model.

The paper assumes wells function without failure. It notes these systems can be deployed almost anywhere with relatively minimal regulatory oversight. As assumptions inside an energy-systems model, that's perfectly reasonable. As assumptions for constructing hundreds of thousands of deep boreholes, they deserve much closer examination.

Because drilling does not happen inside a spreadsheet. It happens underground. And underground keeps its own secrets.

In July of 2024, a drilling crew installed a borehole in the back garden of a home on Cleat Hill, near Bedford, England. The work was ordinary — a closed-loop geothermal bore for a residential heat pump.

About one hundred meters below ground, the drill encountered natural gas. The flow was controlled. The bore was capped. The gas continued to migrate.

One hundred and nine days later, a house exploded. Paul Swales, eighty-five, was killed in the blast. Julia Harris, eighty-four, died from her injuries days later. Roughly fifty families lost access to their homes while investigators secured the area.

The official investigation continues. The sequence of events does not. A geothermal bore encountered an unexpected subsurface hazard. The hazard was not adequately isolated. Gas migrated. People died.

One year earlier, near Orland, California, another drilling crew was constructing a water well. Not a geothermal well. Not an oil well. A water well.

Before sunrise, the bore intersected a pressurized gas zone. The rig caught fire. Fortunately, everyone escaped. The fire could not simply be extinguished because the gas feeding it remained under pressure. Water and gas erupted together while firefighters waited for the formation to exhaust itself.

Different country. Different purpose. Different geology. The same engineering lesson. Unexpected subsurface conditions do not care why the hole is being drilled.

Neither event was an act of God. Neither was inevitable. Both were engineering problems. And engineering problems can be engineered out.

That distinction matters because I am not arguing against geothermal. I am arguing for building geothermal fields with the same rigor we already expect everywhere else in critical infrastructure.

The Chicago paper asks an important question: can geoexchange cool gigawatt-scale data centers? I think the answer is yes.

But deployment introduces a second question. Can we safely construct hundreds of thousands of deep boreholes across widely varying geology while protecting workers, aquifers, neighboring property, and long-term system performance?

Those are not the same engineering problem. The first is a thermal model. The second is construction engineering. One cannot be assumed from the other.

For decades, most geoexchange systems were relatively shallow. A few hundred feet. Typically within or just below freshwater aquifers. They rarely encountered deep gas. They rarely encountered saline formations. They rarely crossed multiple pressure regimes.

That history is encouraging. It is not necessarily predictive.

Today's projects increasingly drill 800, 1,000, and even 1,500 feet. Each additional foot intersects formations that previous generations of geothermal contractors never had to consider.

Using the safety record of shallow residential installations to predict the behavior of industrial-scale deep borefields is like using the accident statistics of a country road to design an interstate highway. The traffic has changed. The engineering must change with it.

Scale changes everything. Six hundred eighty-six boreholes are not simply one borehole repeated 686 times.

Large borefields introduce problems that barely exist in residential work: directional uncertainty, borehole drift, thermal interference, well intersections, barrier verification, cross-formational communication, and construction quality assurance.

None of these invalidate geoexchange. Every one of them requires intentional engineering.

Consider what actually protects groundwater. Every borehole passes through multiple geologic layers — freshwater aquifers, confining clays, deeper formations, and occasionally gas, saline water, or pressurized fluids.

The only thing preventing those layers from communicating with one another is the engineered barrier surrounding the bore. That barrier cannot be assumed. It must be designed. Installed. Verified. Documented.

When Cleat Hill became an emergency, the eventual response relied on oil-and-gas cement plugs specifically engineered to isolate gas-bearing formations. Think about that for a moment. The technology that ultimately controlled the hazard already existed. It simply wasn't part of the original construction standard.

Before any owner authorizes construction of a borefield measured in the hundreds of wells, I believe four questions deserve written answers.

1. Barrier IntegrityHow do you design, monitor, and verify primary and secondary barriers while drilling — and those that permanently isolate one geologic formation from another?
2. Bore PlacementHow do you know where every bore actually went — not where it was intended to go? How is that verified?
3. Bore IntersectionsWhat procedures exist to detect, document, and manage borehole intersections if they occur? Because on sufficiently large projects, they become an engineering problem — not a theoretical one.
4. Thermal PerformanceHow are bore deviations incorporated into long-term thermal modeling so interference between neighboring wells is understood rather than discovered years later?

Those four questions have nothing to do with opposing geothermal. They have everything to do with building geothermal fields worthy of mission-critical infrastructure.

This is exactly why OXBO exists.

R_o-κ evaluates the subsurface before drilling begins. BIOT-κ creates a permanent construction record showing where every bore actually went, how each barrier was installed, and what was verified during construction.

That same philosophy is driving the development of ANSI/IGSHPA/CSA C449. Control the well. Understand the geology before drilling. Verify the barrier. Document the asset. Mission-critical infrastructure deserves construction records — not assumptions.

None of this requires fear. It requires rigor.

The crews at Cleat Hill and Orland were doing honest work using the practices and regulations available to them. The next generation of geothermal construction will operate at a scale those regulations were never written to address.

AI data centers will become some of the most valuable facilities humanity has ever built. Their cooling systems deserve engineering that recognizes drilling for what it is: subsurface construction, with all the uncertainty and responsibility that implies.

The opportunity is real. So are the risks. Fortunately, both are manageable. Ask the questions before drilling begins. Require the answers in writing. Verify the work. Document the result.