Morningstar flags $17.8B cooling market
- Future Market Insights said on May 5 the AI datacenter liquid-cooling market could hit $17.8 billion by 2036 as hyperscale GPU clusters run hotter. - The forecast starts from $3.7 billion in 2026 and implies 16.9% annual growth, with direct-to-chip and immersion systems doing the heavy lifting. - That matters because AI buildouts now hinge on cooling, optics, and power gear — not just GPUs.
Liquid cooling just moved from datacenter plumbing to a real AI bottleneck. A new market forecast out May 5 pegs the AI datacenter liquid-cooling market at $17.8 billion by 2036, up from $3.7 billion in 2026. The basic reason is simple — AI racks are getting too dense and too hot for old air-cooling designs. And that changes what counts as scarce in an AI buildout. (morningstar.com) ### Why is cooling suddenly the story? A normal enterprise server room and a modern AI cluster do not stress infrastructure in the same way. Training and inference racks packed with GPUs push far (morningstar.com) They are no longer niche science-project options — they are becoming baseline infrastructure for high-density AI deployments. (morningstar.com) ### What exactly got forecast? The May 5 note tied the market to hyperscalers, cloud providers, and enterprise AI operators building next-generation compute environments. The headline number is $1(morningstar.com) underneath it — cooling is being modeled as a core layer of AI capex, not a support afterthought. (morningstar.com) ### Why does rack density break air cooling? Air cooling works well when the heat source is spread out and the airflow path stays manageable. AI clusters break both assumptions. Heat is concentrat(morningstar.com)ou are building around dense GPU racks, liquid cooling starts to look less like an upgrade and more like the only practical option. (accessnewswire.com) ### Is this only about cooling hardware? Not really — the catch is that thermal design now collides with the networking stack. DigiTimes flagged a separate squeeze in AI infrastructure this week: the jump from 800G to 1.6T optical modules (accessnewswire.com)lity, packaging complexity, and lead times. Cooling, networking, and power are starting to move as one constraint set. (digitimes.com) ### Where does Applied Digital fit in? Applied Digital’s news is not a cooling announcement, but it shows how companies are reorganizing around AI infrastructure economics. On May 5, the company said it completed the separation of its cloud business into ChronoScale, which is set to trade on Nas(digitimes.com)oint is not the ticker change — it is that infrastructure players are carving out businesses and financing paths around AI demand that now depends on land, power, cooling, and interconnect all at once. (ir.applieddigital.com) ### So what should buyers take from this? If you are budgeting an AI datacenter, the expensive surprise is no longer just GPU availability. It is whether the site can actually deliver power, reject heat, and source the optical gear needed to keep giant clusters fe(ir.applieddigital.com)n forecast. (morningstar.com) ### Bottom line The AI boom is turning cooling into frontline infrastructure. More chips still matter, but the winner may be the operator that can secure coolant loops, power capacity, and optics before everyone else does.
