* This release marks a practical expansion of modular liquid cooling options for AI infrastructure operators facing Vera Rubin power densities
- The 14 MW validated capacity and ETL certification provide the credibility required for production deployment decisions, while the pay-as-you-grow model directly addresses the p...
The Thermal Wall AI Operators Are Hitting
AI data center operators face a brutal arithmetic problem. GPU power densities hit 100+ kW per rack with NVIDIA's Vera Rubin architecture, and air cooling simply cannot keep pace. The industry needed liquid cooling infrastructure that could scale alongside compute deployments without requiring operators to bet the farm on a single massive CDU installation upfront.
The core challenge: most legacy CDU platforms force operators into binary choices. Either deploy a monolithic system sized for theoretical peak load, eating massive CapEx upfront, or piece together under-sized equipment that creates thermal bottlenecks as GPU counts expand. Neither approach works when AI workloads and compute roadmaps shift quarterly.
GigaModular CDU: Modular Building Blocks for Phased Rollouts
LiquidStack, a Trane Technologies subsidiary, addresses this with the commercially released GigaModular CDU platform. The architecture centers on multi-MW modular blocks that operators deploy incrementally, matching cooling capacity to compute growth.
The platform targets the Vera Rubin thermal envelope directly. Each module integrates with centralized control systems designed for coordinated scaling rather than aggregated independent CDU units. This distinction matters: aggregated capacity treats each CDU as a standalone box, while coordinated scaling treats the cooling infrastructure as a unified system aware of load distribution across racks.
Technical parameters:
- Maximum configuration: 14 MW validated per deployment - Silicon compatibility: Supports merchant silicon and hyperscale custom silicon heat profiles - Deployment model: Pay-as-you-grow modular blocks - Application range: AI, HPC, hyperscale data centers - Thermal range: Covers application temperature ranges across silicon generations - Certification: ETL certified following full-load testing
The platform integrates with Trane Technologies' broader thermal management portfolio, giving operators access to global service infrastructure for deployment and lifecycle support. This matters for hyperscale operators running facilities across multiple geographies.
Testing Validation and Market Traction
LiquidStack announced the GigaModular CDU in June 2025. The intervening months involved multi-module system integration testing and full-load validation at the 14 MW configuration. ETL certification confirms the platform meets electrical safety standards for commercial deployment in North America.
The company reports early customer orders reflecting demand for scalable liquid cooling. Given the current AI infrastructure buildout cycle, this tracks with industry analyst projections on GPU density growth through 2026-2027.
"Operators need cooling infrastructure that matches their deployment cadence," the company stated. "Modular architecture removes the CapEx barrier of sizing for theoretical peak load."
The platform's future-ready positioning includes support for next-generation GPU platforms in high-density environments and gigawatt-scale deployments. Whether this proves accurate at scale remains to be seen, but the architecture provides a clearer expansion path than single-system approaches.
---
**
M4S TAKE
My take: certifications like this matter because they give buyers a defensible reason to shortlist a supplier. In a market where everyone claims quality, third-party validation is the difference between being considered and being ignored.
Simon McLoughlin
Is this your company?
This article features your business. Claim it to add your logo, contact details, and a link to your website — or upgrade to reach more buyers.
Did you know 80% of Press Releases trigger AI content warnings? Reach out and the M4S team can assist.