Accelsius has launched the NeuCool MR250, its first row-based Coolant Distribution Unit (CDU), marking a major milestone in the company’s two-phase, direct-to-chip (2P D2C) liquid cooling roadmap. Debuting this week at the Open Compute Project (OCP) Global Summit in San Jose, the MR250 delivers 250 kW of cooling capacity per rack—supporting configurations of either one 250 kW or two 125 kW racks—and targets AI and HPC data centers facing escalating thermal and energy challenges.
The MR250 leverages refrigerant-based liquid cooling to eliminate many of the operational and maintenance burdens of traditional single-phase water systems. It allows facility water temperatures up to 45 °C, extending free-cooling windows and reducing total energy consumption. The system is compatible with environmentally preferred refrigerants such as R1233zd(E) and R-515B, giving operators flexibility to balance performance with sustainability objectives. Accelsius manufactures the units at its newly expanded Austin facility, which has boosted production capacity fivefold to support large-scale deployments starting this quarter and ramping into 2026.
Accelsius and research partners will showcase the MR250 and present multiple OCP sessions covering two-phase micropillar evaporator design and total cost-of-ownership comparisons between single- and two-phase cooling. “The MR250 signals a new chapter for liquid cooling,” said Josh Claman, CEO of Accelsius. “Two-phase, direct-to-chip solutions are no longer just proofs of concept—they’re ready for the scale and reliability that modern data centers demand.”
• Provides 250 kW cooling per rack, scalable to multi-rack configurations
• Enables operation with 45 °C facility water for higher energy efficiency
• Reduces leak and corrosion risk vs. single-phase water systems
• Supports R1233zd(E) and R-515B refrigerants for sustainability goals
• Manufactured in Austin with 500% expanded capacity for high-volume rollout
🌐 Analysis: Accelsius’ NeuCool MR250 demonstrates the growing maturity of two-phase direct-to-chip cooling as AI data centers move toward multi-hundred-kilowatt racks. The company’s partnership with UT Arlington, Celestica, and the University of Maryland highlights an ecosystem approach to validating advanced thermal technologies. Competitors such as CoolIT, Submer, and NVIDIA’s in-house liquid cooling initiatives are also accelerating efforts to meet the thermal limits of next-generation GPU servers and dense AI clusters.
