Legrand Invests in Liquid Cooling for AI Data Centers
Legrand has announced a strategic investment in Accelsius, a specialist in two-phase liquid cooling technology. The move is intended to strengthen Legrand's portfolio of thermal management solutions for the high-density power requirements of AI-ready data center infrastructure.
- Legrand's investment was part of a larger strategy that also included the acquisition of Kratos Industries, a Colorado-based manufacturer of power distribution systems with around $100 million in annual sales. This dual move targets both the power ("gray space") and thermal management ("white space") needs of data centers. - Accelsius specializes in two-phase, direct-to-chip cooling, a method distinct from more common single-phase liquid or air cooling. The technology uses a dielectric refrigerant that boils as it passes over a hot chip; the phase change from liquid to vapor absorbs a massive amount of heat energy, a highly efficient process for managing the intense thermal loads of AI processors. - The push toward advanced liquid cooling is driven by the sheer heat output of new AI hardware. While traditional air cooling becomes ineffective for racks drawing over 25kW, direct-to-chip solutions can handle loads up to 100kW per rack. Next-generation GPUs like Nvidia's are expected to have a Thermal Design Power (TDP) exceeding 1500W. - The investment in Accelsius was part of a $65 million Series B funding round led by Johnson Controls, another major player in building infrastructure. This funding is intended to scale Accelsius's production at its Austin, Texas facility. - Two-phase cooling systems can be retrofitted into existing data centers that are already plumbed for water-based cooling, allowing for incremental upgrades. Accelsius's NeuCool platform uses a non-conductive fluid, which mitigates the risk of catastrophic equipment damage from leaks compared to water-based systems. - In an interesting feedback loop, AI itself is increasingly being used to optimize these complex data center cooling systems. By using predictive analytics and real-time sensor data, AI-driven management software can dynamically adjust cooling to match server workloads, reducing energy consumption by up to 40% in some cases. - The technology Accelsius is developing was originally acquired from Nokia Bell Labs in 2022. The company is also a member of NVIDIA's Inception program, which supports startups revolutionizing industries with technological advancements.