Bengaluru pushes immersion cooling adoption
- Bengaluru-linked operators and suppliers are pushing immersion and other liquid cooling harder as India’s AI data-center buildout collides with rack heat and water stress. - India’s data-center capacity is now projected to hit 10 GW by 2030, while newer AI racks can run past 100 kW each. - That makes cooling a design choice, not a facilities detail — with new demand for fluids, plumbing, service, and local supply chains.
Data-center cooling sounds like plumbing. But in India right now, it is turning into compute strategy. AI racks run much hotter than the server rows most facilities were built for, and cities like Bengaluru are already sensitive to water stress. So the local shift toward liquid and immersion cooling is not really about novelty. It is about fitting more AI into the same footprint without blowing up power and water use. (telecom.economictimes.indiatimes.com) ### What is immersion cooling, exactly? Immersion cooling means putting servers into a sealed bath of non-conductive fluid so the liquid pulls heat away directly. It sits in the same family as direct-to-chip cooling, where liquid runs to the hottest parts without (telecom.economictimes.indiatimes.com)ng for densities beyond 100 kW per rack, and some vendors are designing for much higher. (datacenterdynamics.com) ### Why is Bengaluru part of this story? Because Bengaluru is one of India’s core data-center and enterprise hubs, but it also lives with recurring water anxiety. That combination makes old cooling assumptions look shaky. India’s big data-center clusters are concentrated in met(datacenterdynamics.com)sign process. In other words — the city is not unique, but it is a sharp example of the tradeoff. (datacenterknowledge.com) ### Why is AI making this urgent? Traditional enterprise compute gave operators some room to improvise. AI does not. Training and inference clusters pack more power into each rack, which means more heat in less space. That is why liquid cooling has moved from “nice to have” to “strategic,” as India’s market shifts toward AI-ready campuses and multi-year capacity planning instead of ordinary colocation expansion. (arizton.com) ### Is this already happening in India? Yes — and not just in pilot-lab form. NTT deployed liquid immersion cooling and direct-contact liquid cooling in a production Indian data center back in 2023, saying the setup could improve energy efficiency by about 30%. Infosys and Shell also teamed up in Bengaluru in late 2023 to push immersion-coo(arizton.com)India CDU line and has also been building immersion-related products and coolant partnerships. (datacentremagazine.com) ### So why not just stick with air? Because air cooling starts to look like trying to cool a blast furnace with desk fans. It still has a role, but once densities climb, the energy penalty and space penalty get ugly fast. Liquid systems can cut cooling overhead, support tighter packing, and reduce dependence on evapor(datacentremagazine.com)ng harder to ignore. (datacenterknowledge.com) ### What is the catch? The catch is that liquid cooling swaps one set of headaches for another. Operators need fluids, manifolds, CDUs, leak management, handling protocols, compatible server designs, and technicians who know how to maintain all of it. Immersion also changes hardware service routines and raises questions about fluid sourcing, lifecycle management, and vendor lock-in. That is why local supply chains matter so much here. (datacenterdynamics.com) ### Why does local manufacturing matter now? Because India does not just need more cooling gear. It needs cooling gear that can be deployed, serviced, and replaced locally at AI-build speed. Refroid’s India-made CDU push and its coolant work with Bharat Petroleum show the shape of the next phase — less imported specialty kit, more domes(datacenterdynamics.com)ect details are still thin. (datacenterdynamics.com) ### Bottom line? Bengaluru’s immersion-cooling push is really a preview of India’s AI infrastructure problem. The country wants a lot more domestic compute. But denser racks make cooling a first-order constraint. If operators can make liquid and immersion systems local, serviceable, and water-thrifty, they unlock that growth. If not, the bottleneck stops being chips — and becomes heat.