As AI workloads push data center infrastructure to its limits, Accelsius is betting that the future of cooling won’t just be liquid—it will be fully integrated. The company has launched the NeuCool IR150, a rack-level, two-phase liquid cooling system that combines compute space and cooling infrastructure into a single enclosure capable of handling up to 150kW per rack.
The announcement lands at a critical inflection point for data centers. AI-driven compute density is rising so quickly that traditional air cooling is no longer viable at scale. Even chipmakers like NVIDIA are designing next-generation platforms, including the upcoming Vera Rubin architecture, to rely entirely on liquid cooling.
A Rack That Cools Itself
The NeuCool IR150 is designed as a fully integrated, plug-and-play system, combining:
- A two-phase Coolant Distribution Unit (CDU)
- 42U of IT rack capacity
- Built-in liquid and vapor manifolds
All of it fits into a standard 800mm-wide enclosure, effectively collapsing what would traditionally be multiple systems into one deployable unit.
“The IR150 represents the next evolution of data center infrastructure,” said CEO Josh Claman, pointing to the growing need for simpler, scalable cooling architectures that can keep pace with GPU-heavy AI clusters.
Unlike conventional liquid cooling setups, which often require complex external plumbing and water treatment systems, the IR150 integrates cooling directly into the rack. That reduces deployment time, lowers system complexity, and isolates failures to a single rack—an increasingly important consideration for hyperscale and enterprise AI environments.
Why Two-Phase Cooling Matters
Most liquid-cooled systems today use single-phase water-based cooling, which introduces risks such as leaks, corrosion, and ongoing water quality management. Accelsius takes a different approach with two-phase cooling, using a non-conductive dielectric refrigerant that evaporates and condenses to remove heat.
The benefits are significant:
- Up to 90% reduction in cooling energy consumption compared to air cooling
- Elimination of water usage inside the rack
- Lower risk to sensitive GPU and server components
- Reduced operational complexity and maintenance
Independent analysis suggests Accelsius’ approach can deliver 35–44% annual OpEx savings and 8–17% lower five-year TCO versus single-phase systems—numbers that matter as AI infrastructure costs balloon.
Built for Next-Gen AI Hardware
The IR150 is engineered for both current and future GPU platforms. It supports today’s high-density systems, including NVIDIA Blackwell GPUs, and is designed with enough thermal headroom for next-generation architectures.
It also supports warm-water cooling up to 45°C (ASHRAE W45 standard), enabling more free cooling and reducing reliance on chillers—a key lever for improving data center sustainability.
Other standout features include:
- Hot-swappable components for maintenance without downtime
- Single-rack failure domain for improved resilience
- No water in the rack, eliminating treatment requirements and leak risks
Industry Implications
The launch of the IR150 underscores a broader shift in data center design: cooling is no longer an afterthought—it’s becoming a core architectural layer for AI infrastructure.
As AI models grow larger and more power-hungry, solutions like integrated liquid-cooled racks could become the new baseline, particularly for hyperscalers and enterprises building dedicated AI clusters.
Accelsius is positioning itself squarely in that transition, offering not just cooling technology but a ready-to-deploy infrastructure unit that aligns with how AI systems are increasingly designed—dense, modular, and liquid-cooled from the ground up.
Power Tomorrow’s Intelligence — Build It with TechEdgeAI
