AI and High-Density Data Halls in Texas

About 474 gigawatts of generation and load interconnection requests were sitting in the ERCOT queue by late summer 2026, and roughly 90 percent of them were data centers, according to a Utility Dive report citing Governor Abbott's letter. That number is not a forecast of what will be built. It is a measure of how many developers want power, and it explains why an AI data hall in Texas is a power procurement problem before it is a construction project.

High-density halls for AI training and inference differ from conventional halls in three ways. Each rack draws far more power. Much of that power leaves as heat that air alone cannot remove, so liquid cooling enters the design. And the floor has to hold heavier racks, larger pipes and denser cable tray. We build halls with those facts in mind and take the electrical and mechanical decisions first.

The page that follows describes how we sequence the work, where the schedule risk sits and what an owner should settle before the first foundation goes in.

The same report said that on August 3, 2026 the governor called for a moratorium on new data center interconnections and directed a community impact review of data centers and crypto facilities of 25 MW and above. ERCOT's target for its audit filing is December 10, 2026, and ERCOT said it will not have the Batch Zero study done by its April 9, 2027 deadline. Whatever happens next in policy, an owner planning an AI hall should assume the interconnection date is uncertain and plan the construction.

One account of an AI campus in Abilene describes 1.2 GW of operating capacity, more than 6,400 workers on the campus and current-generation accelerator hardware. That account comes from a secondary source and the operators should confirm it, but it shows what high-density construction at scale looks like on the ground: the account says the campus has become one of the city's largest employers.

We begin with the rack. The power per rack, the cooling method per rack and the weight per rack decide the row layout, the pipe sizes, the electrical busway and the structural loads. Once those are known, the data hall is sized to the number of racks the power can support. Owners often start with a floor plan and discover later that the power or cooling supply cannot fill it, so we reverse the order.

That approach carries into the plant. The chilled water or other cooling supply is sized for the planned heat load, and the generator and UPS systems are sized for the electrical load. Liquid cooling infrastructure coordination covers the piping, the heat exchangers and the testing. The pads for the plant are described on our page about chiller and cooling plant pads.

An AI hall is heavy. Racks loaded with accelerators weigh more than conventional racks, and liquid cooling adds the weight of pipe and fluid. The slab must carry those point loads and stay flat so that racks can be leveled. Our work on mission-critical concrete slabs describes the flatness, curing and joint design that these halls need.

Cleanliness is also tighter. Liquid cooling loops are flushed and filled under controlled conditions, and a single contamination event can damage cold plates. We plan the flushing and filling sequence with the commissioning team and keep pipe protected until it is installed. The building becomes a clean assembly area long before equipment arrives, and we manage who enters and what they carry.

An AI hall seldom stands alone. It depends on a campus-scale electrical service, so the utility and substation site work often has the longest lead. It also depends on the structure around it; for high-density projects we frequently coordinate the shell and core with the plant design so that pipe routes and electrical rooms are built into the building instead of added later.

Owners comparing options should also look at the AI training campus building type, which describes how these halls cluster on a single site. The AI and GPU cloud operators page covers the contract and schedule patterns we see in that market.

We split the schedule into a construction clock and an energization clock. The construction clock is the work we control: site, structure, slabs, rooms and pipe. The energization clock belongs to the utility and ERCOT. Given that interconnection timing is uncertain, we plan the building so that it can be complete and tested on temporary or partial power while waiting for the permanent feed.

Each completed milestone then becomes a decision point. An owner can pause, continue with the building shell, or accelerate the plant work depending on how the interconnection date moves. Photographs and weekly updates help investors follow those decisions in plain language. We also keep a list of long-lead items that can be released early, such as chillers and switchgear, so that the gear is ready when the power is.

Share your rack power target, cooling method and interconnection status. We will return a hall layout that matches the power you can count on, a plant concept and a schedule that treats the utility date as a separate clock. If you are choosing between sites, we can compare them on power and construction terms together.

Do AI halls always need liquid cooling?

Not always, but at the rack densities typical of current accelerators liquid cooling is common, either direct to the chip or through rear-door heat exchangers combined with air. We design the building so that piping, floor loads and plant space are ready, even if the first phase begins with air cooling and liquid is added later.

How does the ERCOT queue change our construction plan?

It changes the order of work. Because interconnection timing is uncertain, we build and test everything that can be built and tested without the permanent feed, and we keep the plant and electrical gear on a schedule that can speed up or slow down. We do not promise an energization date that depends on the utility.

What is special about the slab?

Heavier racks, tighter flatness and more embedded items than a conventional floor. We plan the joints, the reinforcement, the curing and the testing before the pour, and we coordinate sleeves and embeds with the mechanical and electrical designers so the slab is not cut later.

Can we build in phases to match power availability?

Yes. A phased plan lets the owner build the shell and plant for the full load but fit out only the halls the power can serve. We write the phasing so that later halls can be completed without shutting down those already running.

Who tests the liquid cooling system?

The commissioning team does, with our coordination and the mechanical contractor's startup. Pressure tests, flushing, filtration checks and integrated tests under load are sequenced so that no hall is energized until its loop has passed. We keep the records for the owner's file.