\n\n
Key developments: Next‑generation GPUs are now drawing more than 250 kW per rack, making traditional air‑cooling infeasible and pushing the industry toward direct‑to‑chip liquid cooling, which can cut energy use by 30‑60 % and handle power densities above 40‑50 kW per rack. IDC projects that 90 % of large‑scale performance‑intensive deployments will rely on liquid cooling by 2027, and a recent IDC survey found three‑quarters of providers already have AI‑focused racks in production. Major vendors are responding—Carrier has lifted its 2026 data‑center sales forecast to $2 billion and is adding a new HVAC plant, while Trane Technologies is taking pre‑orders for a new LiquidStack coolant distribution unit that will ship in Q2 2027 for NVIDIA’s Vera Rubin GPUs. Schneider Electric and Motivair have also launched an end‑to‑end liquid‑cooling portfolio aimed at the toughest AI‑scale sites.
Why it matters: Liquid cooling not only enables the ultra‑dense AI racks demanded by modern workloads but also reduces fan power and improves PUE, helping operators meet sustainability targets. With power consumption projected to rise from 397 TWh in 2024 to 915 TWh by 2028, the shift to efficient cooling is critical to curb both electricity and water use.
What to watch next: Watch for broader adoption of the new LiquidStack CDU units and Schneider‑Motivair solutions as they become available in 2027, and monitor how carriers like Carrier scale manufacturing capacity to meet surging demand. Continued advances in AI‑driven cooling control—such as reinforcement‑learning optimizers that can cut cooling energy by ~40 %—will likely become a standard feature in next‑generation data centers. [[1, 2, 3]]