SC25 gets heavy with mega power and cooling solutions

Steven

SC25 gets heavy with mega power and cooling solutions

Summary

At SC25 the focus shifted from purely chips and systems to the physical infrastructure needed to run the next generation of AI and HPC deployments. Exhibitors such as Mitsubishi Heavy Industries, Danfoss, Vertiv, Nidec and Eaton displayed large-scale cooling towers, coolant distribution units (CDUs), hydrogen-capable turbines and high-voltage power sidecars in full-scale mockups and dioramas.

The show underlined that AI superclusters demand extraordinary power and thermal management: individual racks are reaching hundreds of kilowatts, entire datacentres can consume gigawatts, and liquid cooling plus specialised DC power distribution are becoming the norm rather than the exception.

Key Points

  • Heavy industrial vendors were prominent on the SC25 show floor, demonstrating that power and cooling are central to modern datacentre builds.
  • High-density racks (NVL72, upcoming Kyber/Rubin designs) are driving widespread adoption of liquid cooling and megawatt-scale CDUs.
  • Hyperscale AI sites (eg OpenAI, Meta) plan facilities consuming multiple gigawatts, far beyond traditional HPC designs like El Capitan.
  • Power distribution is moving from AC to high-voltage DC: vendors showed 800V sidecars and are exploring liquid-cooled 1,500V systems to support 600 kW+ racks.
  • Utilities are a bottleneck; operators are using mobile gas turbine plants and modular/hydrogen-capable power solutions as stopgaps.
  • Thermal management vendors demoed CDUs with capacities from hundreds of kilowatts up to 2+ MW, reflecting growing demand for large-scale heat rejection systems.
  • Given scale and timelines, expect more unusual power solutions (hydrogen turbines, modular plants) — and possibly SMRs — to appear in future shows.

Why should I read this?

Short version: chips are only half the story. If you care about where and how AI gets built, this explains the headaches — and the kit — that solve them. We skimmed the booths so you don’t have to: power, cooling and high-voltage DC are the real rate-limiting steps for big AI datacentres.

Context and Relevance

This article matters because AI scale is changing datacentre economics and engineering. Racks are becoming far denser (120–600 kW+), pushing operators toward liquid cooling, megawatt CDUs and high-voltage DC distribution. That shift affects site selection, utility negotiations, capital planning and vendor ecosystems.

For infrastructure teams and planners, the implications are immediate: expect longer lead times for grid capacity, the need to plan for heat rejection at gigawatt scale, and growing collaboration between traditional heavy-industry vendors and IT OEMs. For suppliers and investors, the market opportunity extends beyond servers to cooling towers, CDUs, turbines and high-voltage power systems — plus potential interest in SMRs and hydrogen solutions as long-term alternatives.

Source

Source: https://go.theregister.com/feed/www.theregister.com/2025/11/20/heavy_industry_invades_sc25/