NVIDIA validates Castrol direct-to-chip coolants for AI factories
The endorsement puts Castrol on a short list for AI cooling specs, but no capital commitments came with it.
NVIDIA has validated two Castrol direct-to-chip cooling fluids for AI factory and data center equipment, putting Castrol's chemistry on a short list of components that can be specified for high-density AI loads. Data Center Dynamics reported the validation, which covers Castrol ON PG25 and PG25T.
Direct-to-chip cooling routes fluid through cold plates mounted on CPUs and GPUs, handling heat levels that air cooling cannot manage. Castrol says the closed-loop design recirculates the fluid without exposing it to the atmosphere, which cuts evaporative water loss and improves water usage effectiveness. That metric matters when water-strained regions weigh data center approvals.
The review went beyond fluid chemistry. NVIDIA examined corrosion resistance, thermal stability, dilution-water quality, propylene glycol base-fluid purity, and compatibility with metals, plastics, and seals. It also weighed Castrol's business continuity plans and global supply-chain resilience, effectively asking whether the company can deliver consistently across markets. Peter Huang, Castrol's global president of thermal management and data centers, called the inclusion in NVIDIA-Validated AI Factory Infrastructure a step toward making liquid cooling a dependable part of AI factory design. Sung A. Kim, the company's data center global technology manager, said the validation confirms the fluids hold up at the scale and intensity of today's AI infrastructure.
For infrastructure investors, this is a design-standard event, not a capital-markets one. No capital stack, no contracts, no capacity numbers, and no offtake were disclosed. The validation suggests the AI cooling supply chain is consolidating around a short list of proven suppliers, and unproven cooling chemistry now faces a narrower procurement path. The supply-chain audit behind the endorsement is part of the procurement equation.