When CRAC Units Become the Bottleneck: Three Cooling Paths Forward

The colocation business is changing faster than most cooling infrastructure was designed to handle. Operators who built their facilities a decade ago designed them for workloads that look nothing like what tenants are asking for today. The CRAC units at the center of those facilities are now the limiting factor — not just in terms of age or efficiency, but in terms of what the facility can actually sell.

Triton Thermal's CRAC replacement guide for colocation operators breaks down the decision into three distinct paths. Understanding those paths — and which one applies — starts with understanding why the status quo is becoming untenable.

The most pressing issue for facilities running CRAC equipment isn't age — it's refrigerant. R-22 production and import ended in 2020. The only supply available today is reclaimed, and costs have risen over one thousand percent from historical norms. Any unit still running R-22 is operating on borrowed time and borrowed money.

R-410A is next. Under the AIM Act, supply drops to thirty percent of baseline by 2029. Starting January twenty twenty-seven, new data center cooling equipment must use refrigerants with a Global Warming Potential below seven hundred. R-410A carries a GWP of two thousand eighty-eight. Replacement refrigerants like R-454B are not compatible with existing equipment. The transition requires full hardware replacement — there is no patch or retrofit that resolves it.

The clock isn't hypothetical. Equipment lead times are already stretching as demand builds industry-wide. Vertiv's current order backlog is nearly ten billion dollars. Operators who start planning in twenty twenty-seven will be competing with everyone else who also waited.

Refrigerant regulations are a compliance problem with a fixed deadline. The density problem is already costing operators business. NVIDIA's H100 generation requires approximately forty kilowatts per rack. The GB200 demands over one hundred twenty. Standard CRAC-based colocation facilities were built for three to eight kilowatts per rack. Water moves heat roughly three thousand five hundred times more efficiently than air — at fifty kilowatts per rack, air cooling would require nearly eight thousand cubic feet per minute of airflow. That's a physical impossibility in a standard raised-floor environment.

The Triton Thermal team frames the replacement decision around three paths, each with different cost profiles, timelines, and strategic positioning.

Path one is like-for-like CRAC replacement. New equipment, compliant refrigerants, modest efficiency gains of twenty to thirty percent. It solves the regulatory problem and improves PUE modestly, but the density ceiling stays around fifteen to twenty-five kilowatts per rack. It's the right call for operators with newer facilities and tenants that don't need AI-scale density — but it doesn't reposition the facility for the next decade.

Path two is conversion to centralized chilled water with CRAH air handlers. PUE typically drops from the one-point-seven to two-point-zero range down to one-point-four to one-point-five. Waterside economizers unlock hundreds of free cooling hours per year. A documented case study showed PUE dropping from one-point-eight to one-point-three after a full CRAC-to-CRAH conversion, translating directly into sustained operating cost reduction.

Path three is transition to hybrid or full liquid cooling — the path for operators competing for AI and HPC workloads. Direct-to-chip cold plates, rear-door heat exchangers, and immersion cooling support thirty to over two hundred kilowatts per rack. PUE approaches one-point-zero-three to one-point-two. The addressable market — AI infrastructure tenants actively searching for high-density colocation — commands premium rates.

Most operators deploying new cooling capacity aren't choosing between air and liquid in absolute terms. The practical approach is zoned: CRAH-based air cooling for standard enterprise workloads, rear-door heat exchangers for moderate-density racks, and direct-to-chip liquid cooling for the highest-density GPU deployments.

The facilities that compete for the next generation of AI and HPC tenants will be the ones that made cooling infrastructure decisions in twenty twenty-five and twenty twenty-six — not the ones who waited for the 2029 refrigerant cliff to make the decision for them.

This content was developed with the support of Houston digital marketing agency ASTOUNDZ.

Triton Thermal
City: Houston
Address: 3350 Yale St.
Website: https://tritonthermal.com/
Phone: +1 832 328 1010
Email: marketing@hts.com