Accelsius’ first IR150 install puts two-phase cooling into data center training playbooks

EdgeConneX is putting two-phase liquid cooling into the place where data center habits are formed: the training lab.

On Aug. 27, EdgeConneX said it is deploying Accelsius’ IR150 at its Global Training Center, a facility built to standardize how teams operate data centers before they touch customer environments, according to Facilities Dive. Accelsius called it the first commercial deployment of the IR150, which the company positions as a fully integrated IT rack plus cooling distribution unit (CDU) aimed at plug-and-play two-phase, direct-to-chip cooling.

For facilities directors, data center ops leaders, and infrastructure procurement teams, the operational implication is less about a single rack and more about what it takes to make warm-water cooling repeatable at scale: documentation, commissioning steps, spare parts, and technicians who have already seen the failure modes in a controlled environment.

Warm-water setpoints are becoming a spec item, not a nice-to-have

Accelsius’ pitch is that two-phase direct-to-chip cooling can hold GPU temperatures down even as operators push facility-water temperatures up. In results it released July 21, Accelsius said independent third-party benchmarks on a commercially available eight-way NVIDIA B200 GPU server covered roughly 40,000 operating points under simulated AI workloads, and showed 9°C lower B200 junction temperatures versus the server’s factory-installed single-phase direct-to-chip cooling at warm-water conditions, according to Business Wire.

The same release said single-phase performance at 45°C facility water could be matched using Accelsius’ system with 54°C inlet water. It also put a simple energy heuristic on the table: every 1°C increase in facility-water temperature corresponds to about 4% annual energy savings. If a team is modeling a move from 45°C to 54°C, that framing turns a mechanical decision into a financial conversation.

The Register, which separately covered the same benchmark narrative on July 21, described the warm-water angle in practical plant terms. Warmer facility water can reduce reliance on chillers and expand where “free cooling” is feasible, and it can also change the heat-rejection hardware mix, including the balance between dry coolers and evaporative approaches, according to The Register.

The adoption bottleneck may be the CDU, not the coolant

EdgeConneX’s training-center install matters because two-phase is not a “swap the fluid and go” exercise. The Register reported that Accelsius’ approach still looks similar to conventional liquid cooling at the rack level, with manifolds, cold plates and a CDU exchanging heat to facility water. The difference is the working fluid and the way vaporization improves heat transfer at the chip.

But two-phase generally requires CDUs designed for two-phase operation, and the B200 benchmark system described by The Register used custom cold plates and a CDU built for two-phase. That is the operational trapdoor for buyers: a server SKU might look familiar, but the CDU and plate ecosystem defines what can be standardized across halls, and what becomes a one-off integration project.

This is where Facilities Dive’s detail about EdgeConneX’s training mission becomes more than marketing. If a training center is meant to simulate production operations, it is also the place to pressure-test service procedures: coolant handling, leak response, sensor calibration, and how the CDU behaves during power events or flow changes.

What the benchmarks change in capital planning and site design

Accelsius’ released benchmark claims are tuned to a hardware reality data center operators can’t ignore. The Register noted that NVIDIA rack systems can reach system power levels around 250 kW this year, with even higher density on the roadmap. At that point, the cooling conversation stops being about a “liquid loop” and becomes a site-level design question: heat rejection capacity, redundancy scheme, and what facility-water temperatures are sustainable in the local climate without adding mechanical complexity.

In Accelsius’ own announcement, the company pointed to a key variable facilities teams already watch closely: flow rate. Business Wire reported that the release said the thermal advantage versus single-phase cooling rose from 9°C to 14°C when the facility-water flow rate was increased, moving from 1.5 LPM/kW up to 3.0 LPM/kW. The same statement said facility-water supply temperatures could reach 59°C. Changes of that kind carry implications for pump sizing, piping, valve selection, and the hydronic-side energy modeling of the plant.

Where this lands for teams writing specs in 2026

• Ask integrators to quote the CDU as a first-class system, not an accessory. Confirm maximum supply temperature, allowable flow ranges, redundancy model, and how faults are detected and alarmed at the BMS/DCIM layer, since two-phase may require specialized CDU designs (The Register).
• Translate warm-water targets into an energy and mechanical scope change. Accelsius’ published heuristic of roughly 4% annual energy savings per 1°C increase in facility-water temperature is a useful, citable way to run sensitivity analysis in chiller and heat-rejection decisions (Business Wire).
• Use training and serviceability as a vendor filter. EdgeConneX choosing to deploy IR150 in a production-like training environment is a reminder to require documented commissioning steps, technician training, and defined spares for any high-density liquid-cooled rollout (Facilities Dive).
• If higher flow is part of the plan, treat it like a plant redesign. Accelsius’ figures on moving from 1.5 to 3.0 LPM/kW imply downstream changes to pumps, pipe sizes, and control valves that should be priced and scheduled early, not discovered during fit-out (Business Wire).