Accelsius supplies two-phase cooling rack to NSF efficiency research center at Villanova
Summary
Accelsius, an Austin, Texas-based supplier of two-phase, direct-to-chip liquid cooling for AI and high-performance computing, said on Nov. 3, 2026 that it has joined the National Science Foundation's Center for Energy-Smart Electronic Systems (ES2). The center is a federal university-industry research program whose goal is to raise U.S. data center energy efficiency by 20–35%.
As part of its participation, Accelsius will place its NeuCool IR150 at Villanova University, a founding member of ES2. The unit will serve as shared two-phase cooling infrastructure for the program's system-level testing. Accelsius describes the IR150 as the first product in the industry to fully integrate an IT rack with a coolant distribution unit (CDU).
HPCwire carried the announcement, which also notes that Accelsius is a member of the NVIDIA Inception program. The source text is truncated. It does not give the value of the contribution, the length of the arrangement or the specific research projects the equipment will support.
Why it matters
Liquid cooling is becoming a prerequisite for dense AI racks, and two-phase direct-to-chip designs are competing with single-phase systems for that role. Placing an integrated rack-plus-CDU unit inside a federally backed research center gives operators and hyperscale customers a route to independent, system-level performance data on two-phase cooling. That kind of evidence can shape cooling procurement decisions and efficiency targets. The program's stated 20–35% efficiency goal also ties cooling choices directly to power consumption, which remains a central constraint in data center siting and capacity planning.
Expert Lens
Two-phase cooling has struggled less with thermal performance than with buyer confidence in fluid handling, maintenance practices and long-term reliability at fleet scale. Neutral academic testing may help ease those concerns, but research rigs rarely reproduce the operating conditions of a production AI hall. An integrated rack-plus-CDU design could simplify deployment, yet it may also narrow operators' options for mixing vendors across racks and facility water loops. One signal to watch is whether ES2 publishes comparative efficiency results for two-phase and single-phase systems under matched AI workloads.
Sources
This brief was drafted with AI assistance from public sources; every number is attributed to the sources above. See the AI-assisted Disclosure and Methodology.