Google cooling initiative could reshape AI data-centre retrofit market- Bernstein

Google recent unveiling of 'Brazos,' an open-source liquid-to-air cooling design for AI data centers within the Open Compute Project (OCP) ecosystem, is poised to shake up the lucrative AI data-centre retrofit market, according to Bernstein analysts. This move could accelerate the commoditization of some cooling equipment, particularly for AI inference workloads in existing facilities, rather than the more demanding AI training systems. While not an immediate threat to giants like Vertiv and nVent, the initiative signals a shift towards standardized, less complex cooling solutions. The broader context is the explosive demand for AI, which is pushing data centers to their thermal limits, as traditional air cooling struggles with rack densities soaring past 100 kilowatts and heading towards 300 kilowatts and beyond. Liquid cooling, including direct-to-chip and immersion cooling, has emerged as an essential, not optional, solution, offering significantly higher efficiency and reduced energy and water consumption. Retrofitting existing data centers with these advanced cooling systems is becoming a critical strategy, offering up to 80% cost savings compared to building new greenfield sites, especially as new construction faces increasing delays due to land, power grid connections, and permitting hurdles. This development suggests hyperscalers might increasingly prioritize upgrading existing infrastructure over new builds, offering a faster path to bringing AI capacity online. However, the commoditization risk for simpler cooling units like Brazos means established suppliers may need to adapt their strategies to maintain margins, while the overall data center cooling market is still projected to grow significantly, driven by the insatiable demand for AI. The industry will be watching closely to see how this balance of innovation, standardization, and market dynamics plays out in the race to cool the AI future.