Accelsius expanded the power capacity of its Springdale lab and facility in Austin, Texas, to support multi-megawatt, two-phase liquid cooling testing for next-generation AI racks. The site is intended for full-scale validation of Accelsius’ NeuCool two-phase, direct-to-chip platform at higher rack power levels and warmer facility water temperatures.
The company says the upgraded environment is designed to test at “megawatt-class rack densities,” including warm-water conditions. Accelsius also plans to expand its lease footprint at the same location by an additional 18,000 square feet in January 2027.
Accelsius frames the lab upgrade around higher-density GPU infrastructure, citing publicly disclosed GPU roadmaps pointing to 600 kW racks in 2027, alongside megawatt-class rack and pod designs. With multi-megawatt capacity at the Springdale site, the company says it can power a single rack at megawatt-plus loads, or run multiple racks at today’s highest densities, while characterizing end-to-end two-phase cooling performance “from cold plate to coolant distribution unit (CDU) to facility loop” under real-world conditions.
Warm-water testing matters because the allowable facility water temperature drives how hard (and how efficiently) a data center cooling plant can run. If two-phase direct-to-chip systems can hold junction temperatures in check at higher inlet temperatures, operators have more room to reduce chiller hours or run higher chilled-water temperatures, but those gains only count if they hold up at full rack and row scale.
“Two-phase cooling has moved from proving the physics to scaling the product, and that means testing at the power levels our customers are actually planning for,” said John Hewitt, CEO of Accelsius. “With multi-megawatt capacity at Springdale, we can run full racks at megawatt-class loads on warm facility water and show operators exactly how NeuCool performs under the conditions we expect their next AI factories will face.”
Accelsius also pointed to prior benchmark results it says were published in July: NeuCool delivered 9° C lower NVIDIA B200 junction temperatures than single-phase direct-to-chip cooling in warm-water conditions, and matched single-phase performance at NVIDIA’s 45° C facility water benchmark with inlet water as warm as 54° C. Accelsius also claims that each 1° C increase in facility water temperature is worth approximately 4 percent in annual energy savings.
The company said the upgraded lab will support its NeuCool HyperStart program, which it described as a structured path for hyperscale operators, high performance compute providers, and partners to validate two-phase, direct-to-chip cooling and incorporate it into reference designs and roadmaps. Accelsius said several hyperscalers are engaged in the program.
Separately, Accelsius said it is expanding its NeuCool multi-rack family with a 375 kW-capable unit as the next step beyond the MR250, with additional details planned “in the coming months.”
Source: Accelsius
