Liquid Cooling Revolutionizes AI Infrastructure: AMD and NVIDIA Lead Industry Shift

August 17, 2026
Liquid Cooling Revolutionizes AI Infrastructure: AMD and NVIDIA Lead Industry Shift
  • TrendForce notes that cloud service providers, including Chinese operators, are upgrading AI data center architectures, aligning with AMD’s Helios and MI450/MI500 line as liquid cooling gains traction.

  • Institutional investors adjusted AI holdings in the second quarter, with notable moves among NVIDIA, AMD, and Broadcom, while adding SpaceX and Cerebras.

  • Cooling expansions extend beyond GPUs to include NICs, busbars, transceivers, and other rack components as thermal demands rise.

  • The push for higher compute density and tens to hundreds of kilowatts per rack makes cooling and thermal management a strategic priority.

  • Liquid cooling is becoming standard for high-end AI infrastructure, with adoption among AI chips projected to reach about 60% by 2027, up from 53% in 2026.

  • AMD is expanding its AI strategy with the Helios rack-scale solution and continues developing its MI450 and MI500 platforms to compete with NVIDIA, with large-scale shipments anticipated in 2027.

  • Dell PowerCool enables up to fourfold higher GPU density and up to 160 kW per rack via rear-door heat exchangers, cutting cooling energy use by as much as 74%.

  • NVIDIA faced a Rubin heat spreader design decision, moving from a two-piece plan to a one-piece design due to substrate warpage, with Jentech remaining the sole Rubin spreader supplier.

  • Microsoft is deploying zonal and direct liquid cooling for NVIDIA GB300 NVL72 systems, achieving roughly 136 kW per rack.

  • Liquid cooling systems use cold plates, manifolds, and CDUs, with NVIDIA’s Vera Rubin platform expanding to a fanless all-liquid-cooled architecture and adding components like CX9 cards and power boards.

  • Major cold-plate suppliers include Cooler Master, AVC, BOYD, and Auras, with AVC planning Rubin cold-plates shipments and participation in AI ASIC supply chains for AWS, Google, Meta, and OpenAI.

  • Google leads in liquid cooling adoption, applying it to more than 80% of its AI servers and expanding from server to rack-scale cooling as TPU power grows, while NVIDIA and AMD drive broader industry shift toward Vera Rubin and related solutions.

Summary based on 4 sources


Get a daily email with more AI stories

More Stories