Preview. Texas expects up to 78 GW of data-center load and up to 161 billion gallons of annual water demand by 2030 — while water is becoming a critical resource. Our lab-validated CRIOJET cooling — EER 5–6, no compressors, no freon, zero evaporative water — points to a PUE of ~1.2 in the Texas climate. The numbers, and the pilot we propose, below.
Our colleagues from Texas reached out with a specific question: is it possible to cool a data center efficiently and without an enormous water draw — because water in the state is now a problem. We answer with numbers.
⚡ The scale of the problem: electricity
According to ERCOT forecasts, data-center load in Texas could reach 78 GW by 2030 — against 29.6 GW in the previous year’s forecast; grid peak demand is projected at 139 GW by 2030. Even if only half of the applications materialize, that is tens of gigawatts of new load, and 30–40% of every megawatt is not computing but cooling. The global average PUE holds at 1.5–1.6: for every 1 MW of useful IT load, the operator buys 0.5–0.6 MW of overhead, most of it for heat removal.
💧 The scale of the problem: water
Operating data centers in Texas already consume about 25 billion gallons of water per year (about 95 million m³). Estimates for 2030 range from 29 to 161 billion gallons per year — up to 2.7% of the state’s total water use — and all of this at the very moment when water is becoming a critical resource in ever-shorter supply, as we have written in previous articles (link ___). The cause is evaporative cooling: the typical WUE of such systems is around 1.8 L/kWh. For a 1 MW IT hall, that means 8,760 MWh of heat per year and, accordingly, about 15.8 million liters of evaporated water per year — 4.2 million gallons per megawatt. A 100 MW campus evaporates the annual consumption of a town of 10,000–15,000 residents. Municipalities are already responding: permits for new sites are increasingly conditioned on water terms or rejected outright.
The alternatives are known, and both are bad for Texas. Dry chillers consume no water, but at +40 °C (104 °F) ambient their energy consumption grows by tens of percent and PUE climbs to 1.5 and above. Free cooling works only a small part of the year in the Texas climate. Liquid cooling lowers the temperature at the chip, but the heat still has to leave the building — the final loop runs into a cooling tower or a dry cooler again.
The price of these alternatives is measurable. A dry cooler at +40 °C (104 °F) ambient operates with an 8–12 K (14–22 °F) approach to the coolant temperature: the fan fields consume 8–12% of the rejected thermal power, and more in peak hours — exactly when ERCOT is most expensive. Immersion cooling requires dielectric fluids costing tens of dollars per liter and a complete replacement of server mechanics — for operating halls this is not a retrofit but new construction.
🔬 What CRIOJET is: the status, honestly
CRIOJET is ARBOK’s freon-free cooling system built on vacuum coolers, tangential aerodynamics, and phase transitions. Cooling is generated by rapid evaporation of a working fluid — water or safe alcohols — in a low-pressure zone with latent-heat extraction. No compressor. No freon. The working fluid circulates in a closed loop through a condenser-separator that recovers the phase-transition energy: no evaporative discharge of water into the atmosphere, no cooling towers.
It is important to fix the status upfront: a completed stage of R&D calculations and all laboratory experiments, which delivered results; the technology stands at TRL 5 — validation at bench level in a relevant environment. All figures below are bench measurements, with potential for further improvement.
The parameters showed: a CRIOJET module cools 1,000 m³ of air from 50 °C (122 °F) to 10 °C (50 °F) in 8–10 minutes, consuming 2.5–3 kWh for the process — about 400 m³ of cooled air per 1 kWh at a 40 K (72 °F) delta. EER: 5–6. Ambient operating range: from –10 °C to +56 °C (14 °F to 133 °F) — the Texas +43 °C (109 °F) of July sits inside the range with margin. Control is autonomous, PLC-based.
📊 Translating to a data center
Cooling 1,000 m³ of air by 40 K (72 °F) removes ~13.4 kWh of heat. Doing it in 8–10 minutes means 80–100 kW of thermal power per module. Hence the engineering arithmetic of a hall:
A 1 MW IT load requires 10–13 modules in continuous rotation; N+2 redundancy means 12–15 modules, not a second multi-million-dollar chiller. The failure of one module removes 8–10% of cooling capacity, not 50%.
Annual heat rejection for 1 MW is 8,760 MWh. At an effective coefficient of 4.5–5.4, this takes 1,600–1,950 MWh of electricity per year: a cooling overhead of 0.18–0.22, a calculated PUE of ~1.2 — in the Texas climate, without water.
CRIOJET operates automatically: the system switches on by itself to maintain the required temperature. In Texas the thermal load is constant — outside from the climate, inside from IT equipment — so the modules run in continuous rotation year-round. The economics rest not on rare start-ups but on the cost of each cycle: EER 5–6 against COP 1.5–2.5 of classical vapor-compression systems, plus the TSM envelope cutting external heat gains 3–4× — the cooling pays only for the IT load, not for the Texas sun.
♻️ Water: the main answer to Texas
An evaporative system per 1 MW IT: ~15.8 million L/year into the atmosphere. CRIOJET: 0 L of evaporative discharge — the working fluid stays in a closed loop. A 100 MW campus: savings of ~1.6 billion liters of water per year, about 420 million gallons. Against the state forecast of 29–161 billion gallons, every gigawatt built without evaporative cooling removes about 4.2 billion gallons per year from the Texas water balance.
The context is aggravated by the fact that the current Texas water plan does not account for data-center growth: the burden fell on municipalities, each solving the issue its own way — from raised tariffs to outright denials. For an operator this means water has turned from a cost line into a risk factor for obtaining the permit itself: a project with zero evaporative water draw passes the same procedure faster and cheaper.
💰 The money
Electricity. At a PUE of 1.5–1.6, a megawatt of computing drags 3,500–4,500 MWh of overhead per year; at the calculated PUE of 1.2 — 1,600–1,950 MWh. The delta is 1,800–2,500 MWh/year per 1 MW. At $0.12/kWh this is $215,000–300,000 of savings per megawatt per year.
Water. 4.2 million gallons per megawatt per year at a water-and-sewer price of $5–10 per 1,000 gallons — another $21,000–42,000/year, not counting the cost of water rights and water treatment.
CapEx. A classical refrigeration plant — chillers, cooling towers, pump rooms, water treatment — typically takes 15–20% of a data center’s construction cost. The CRIOJET architecture is an array of nozzle blocks on standard industrial power: no compressor hall, no refrigerant licensing chain, no water-treatment station. Scaling from 1 to 5 MW means adding modules, not redesigning a central plant.
In total, for a 5 MW IT site: $1.1–1.5 million/year on electricity, $105,000–210,000/year on water, minus the CapEx for the chiller plant and cooling towers at the construction stage, minus the service budget of the compressor fleet. The calculated payback of the modular array is 1–2 years, by analogy with the calculated cold-storage case; the exact figure will come from the pilot.
Maintenance. The system contains no complex machinery — mainly evaporation tanks. There are none of the overhauls that form the main line of the refrigeration service budget in the classical scheme. The CRIOJET schedule per the technology’s base card: periodic vacuum-leak inspection, annual service of the vacuum pumps, no consumables. Two staff refrigeration technicians handle it after brief training — no separate round-the-clock shift of refrigeration-plant operators is needed.
Compliance. No freons on site — F-Gas regulations and equivalent EPA programs do not apply to the facility. No evaporative water draw — water objections at permitting are removed at the design level.
📈 ERCOT peaks: an additional effect of the design
For Texas, what matters is not only annual energy but behavior in peak hours, when ERCOT wholesale prices hit the $5,000/MWh cap and transmission charges are tied to the four annual peaks (4CP). Two measured properties work together here: cooling speed (a full cycle in 8–10 minutes, response to a temperature excursion without spinning up a compressor cycle) and the TSM envelope slowing heat gains 3–4×. This opens a mode that we so far declare as a design option for the pilot, not a measured fact: pre-cooling the hall by 2–3 K (3.6–5.4 °F) before the peak window and holding the temperature with the building’s thermal inertia for 15–40 minutes with cooling load shed. For a 5 MW IT site this is 0.9–1.1 MW of controllable load in the peak window — an asset ERCOT pays for rather than penalizes.
🏭 The building envelope: the same calculated case
Earlier we published a calculation for a 100,000 m³ cold-storage facility: CRIOJET plus the TSM thermal barrier (a ceramic coating with thermal conductivity of 0.001 W/m·°C; 3 mm performs like 14–18 cm of foam, heat gains through the envelope drop 3–4× — link ____) cut annual cooling consumption from 2,250,000 to 150,000–200,000 kWh. For a data center, TSM solves a different task: it deletes the external climate from the thermal balance. The solar load on a roof in Texas is hundreds of kilowatts for a building of that size; after TSM, the cooling system works only against the internal IT load. A side conclusion for property owners: a TSM-treated warehouse with an industrial grid connection is a ready-made shell for a 1–5 MW edge hall, without the queue for a new grid connection that in ERCOT has stretched into years.
🎯 What we propose to Texas
Not a promise but a measurable pilot: one hall of 0.5–1 MW IT load, an array of 6–13 CRIOJET modules, 12 months of continuous operation through the summer peak, measured values — PUE, WUE, response time to thermal excursions, parameter degradation, module failure statistics — side by side with a classical baseline on the same site. The pilot’s instrumentation is standard: separate electricity meters on the IT buses and the cooling loop (PUE per ISO/IEC 30134-2), flow meters on loop make-up (WUE per ISO/IEC 30134-9), thermography of the envelope before and after TSM, rack-level temperature logging at 1-minute intervals. We open the data to the partner in real time.
The bench data say we will see a PUE of ~1.2 and a WUE of 0. The pilot will either confirm this in an operational environment or show where the calculation diverges from practice — both versions of the result are worth more to the industry than another marketing promise.
Texas is choosing between 161 billion gallons of water per year and a change in the physics of cooling. We propose starting with one measured megawatt.
👉 Working on data-center capacity or site water balance in Texas and the Southwest? DM me — ARBOK is structuring a CRIOJET pilot track.
Also published on: LinkedIn — Michael Vischmidt
