The conventional architecture for liquid-cooled deployments separates the coolant distribution unit from the server rack. A floor-standing CDU sits at the end of a row or in a dedicated mechanical room. Manifold headers run above or below the rack row. Server-side cold plates connect to those headers through quick-disconnect fittings. The CDU conditions the fluid; the rack holds the compute. Two pieces of infrastructure. Two vendors. Two maintenance contracts. Two failure domains that must be managed together at the facility level.
Accelsius announced general availability of the NeuCool IR150 at Data Center World 2026 on April 20, 2026. The product collapses that architecture. It integrates a two-phase coolant distribution unit, 42U of IT rack space, and all internal manifolding into a single 800mm-wide enclosure. The CDU is not beside the rack. It is the rack.
The 800mm footprint allocates 200mm to the integrated cooling infrastructure and 600mm to the IT enclosure. That 600mm column holds 42U of standard server space, meaning operators lose no usable rack units compared to a conventional 600mm rack, only floor space for the additional 200mm width. The cooling capacity is rated at 150 kW per unit — enough headroom for current-generation NVIDIA Blackwell configurations and, according to Accelsius, thermal headroom designed to accommodate NVIDIA Vera Rubin without a hardware change.
The refrigerant is a non-conductive dielectric fluid with an A1 safety rating and low global warming potential. Non-conductive is the operative word for where the fluid lives relative to the IT equipment. The two-phase loop runs inside the cooling column. The IT enclosure stays dry. There is no water or coolant inside the server space. If something fails in the cooling circuit, the servers do not see liquid. That is a fundamentally different risk profile from architectures where coolant lines run directly over or through the server chassis.
The system is ASHRAE W45 compatible, meaning it can accept facility water supply at temperatures up to 45°C. Most data center facilities operate chilled water loops significantly colder than that. W45 compatibility means operators with warm-water or economizer-heavy cooling plants can run the IR150 without chilling the supply loop down to traditional 7°C or 12°C temperatures. That matters for total facility efficiency because chiller energy consumption scales sharply with the temperature differential being maintained.
The architecture creates what Accelsius describes as a single-rack failure domain. In a conventional row-level CDU deployment, a CDU failure affects every rack on that distribution circuit — potentially an entire row of compute. The IR150 contains its cooling infrastructure within each unit. A cooling system fault is bounded to one rack. The racks beside it continue operating on their own independent loops.
This changes the risk calculation for operators running mixed workloads or multi-tenant environments where a row-level thermal event is unacceptable. It also changes how operators think about maintenance windows. The IR150 ships with hot-swappable pumps, power supplies, sensors, the control board, and the LCD display. Maintenance that would otherwise require a planned cooling outage can be performed under live load.
Single-phase direct-to-chip cooling — water or water-glycol running directly through cold plates — is the dominant liquid cooling modality in current deployments. It is well-understood, uses familiar fluid handling equipment, and works with existing ASHRAE water treatment protocols. The IR150 does not use single-phase. It uses two-phase refrigerant-based cooling, where the working fluid undergoes liquid-to-vapor phase change as it absorbs heat at the cold plate, then returns as liquid after rejecting heat at the facility-side heat exchanger.
The phase-change mechanism carries substantially more heat per unit of mass flow than single-phase convection. Accelsius states 35 to 44 percent annual OpEx savings versus single-phase direct-to-chip systems and up to 90 percent cooling energy reduction compared to air cooling. The 90 percent figure is relative to air cooling's baseline fan and CRAC energy consumption; the 35 to 44 percent figure is the more operationally relevant comparison for operators evaluating liquid cooling modalities against each other.
The IR150 also eliminates water treatment requirements. Single-phase water systems require ongoing inhibitor management, conductivity monitoring, and periodic flushing to prevent corrosion and biological growth in the coolant loop. The dielectric refrigerant used in the IR150 does not require that maintenance regime. For operators managing large fleets, that is a non-trivial operational simplification.
Alongside the IR150 general availability announcement, Accelsius launched the NeuCool HyperStart program at Data Center World 2026. The program is designed to accelerate customer deployments by compressing the evaluation-to-production timeline. Details of specific program terms were not fully disclosed at announcement, but the framing positions it as a structured onboarding path for operators looking to move from pilot to production-scale two-phase deployment without building internal expertise from scratch. For operators who have watched liquid cooling evaluations stall at the pilot stage, a vendor-structured path to production is a more actionable offer than hardware availability alone.
Two-phase direct-to-chip cooling has been the subject of serious interest and limited deployment for several years. The technical arguments for it — superior heat transfer coefficient, lower mass flow requirements, elimination of water treatment — have been well-understood for a long time. The deployment friction has been equally well-understood: unfamiliarity with refrigerant handling, limited OEM validation, and the absence of a standardized installation base that facility technicians could be trained against.
A pre-integrated, factory-tested unit that ships as a complete system and accepts facility water at W45 temperatures addresses the installation familiarity problem directly. There is no refrigerant circuit to commission on-site. The system arrives pre-charged and tested. The facility connection is a water supply and return, which every data center mechanical team already knows how to handle.
Accelsius raised a $65 million Series B in January 2026, led by Johnson Controls with Legrand as a strategic investor. Johnson Controls runs the Silent-Aire CDU platform at 500 kW to 10 MW capacity and the YORK YVAM magnetic bearing chiller line. Their lead investment is not a passive financial bet. It is a strategic position in a competing cooling architecture from the company that also sells the facility-side infrastructure that alternative architectures would replace or reduce. The IR150 general availability is the product delivery that follows that capital.