Key takeaways

  • The Trane‑Eaton reference design promises up to 15 % better energy efficiency, 80 % less copper, and 30 % lower installation costs versus conventional low‑voltage builds.
  • The design is delivered through Trane’s Continuum Rubin DSX and Eaton’s Beam Rubin DSX platforms and follows NVIDIA’s DSX blueprint.
  • Castrol’s ON PG25 and PG25T fluids have earned NVIDIA validation for direct‑to‑chip cooling in AI factories, after a technical review of performance, material compatibility, and supply‑chain resilience.
  • Both projects focus on improving heat‑removal and power‑delivery efficiency for high‑density AI workloads.

Integrated power‑and‑cooling reference design from Trane and Eaton

Trane Technologies and Eaton announced a joint reference architecture that merges medium‑voltage power distribution with advanced thermal‑management modules. The solution is positioned as a repeatable stack that data‑center operators can deploy for AI‑focused facilities.

According to the vendors, the combined system can achieve energy‑efficiency improvements of up to 15 %, while reducing copper usage by as much as 80 % and cutting installation expenses by roughly a third when measured against traditional low‑voltage configurations. Those percentages stem from the companies’ own modelling for high‑density AI sites.

Both platforms – Trane’s Continuum Rubin DSX and Eaton’s Beam Rubin DSX – are built to cooperate with NVIDIA’s DSX ecosystem, including the Omniverse DSX Blueprint. By aligning power and cooling controls, the architecture offers a coordinated approach that simplifies deployment and reduces the risk of mismatched electrical‑thermal designs.

“AI and high‑performance computing are transforming the demands placed on data centers, and customers want solutions that can keep pace with their needs,” said Mauro J. Atalla, senior vice‑president and chief technology and sustainability officer at Trane Technologies.

“By combining our advanced thermal management solutions with Eaton’s innovative power management solutions, we’re delivering a coordinated design that helps customers accelerate deployment, improve efficiency and confidently plan to scale for the future,” Atalla added.

Eaton’s senior vice‑president and chief technology officer for the electrical sector, Michael Regelski, emphasized the speed benefit: “We’re advancing the industry standard for speed of deployment by progressing reference designs into unified systems teams can deploy repeatedly.”

NVIDIA’s AI‑infrastructure lead, Vladimir Troy, framed the partnership in terms of system‑level coordination: “AI factories demand tightly coordinated power, cooling and compute infrastructure to operate efficiently at scale. This new integrated approach supports the robust, scalable foundations enterprises need to unlock the full potential of generative and reasoning AI and turn data into faster, smarter outcomes.”

Summary of claimed efficiencies

MetricConventional low‑voltage designTrane/Eaton integrated design
Energy‑efficiency gainBaseline+15 %
Copper usageBaseline‑80 %
Installation costBaseline‑30 %

All three percentages come from the joint announcement and reflect the projected savings when the reference design is applied to AI‑grade compute clusters.

Castrol ON PG25 and PG25T – NVIDIA‑validated direct‑to‑chip fluids

Castrol announced that its two direct‑to‑chip cooling fluids – ON PG25 (an outside‑air‑temperature formulation) and PG25T (an intake‑air‑temperature formulation) – have passed NVIDIA’s validation process for use in AI factories and data‑centre infrastructure.

The validation required a technical review that covered corrosion resistance, thermal stability, dilution‑water quality, propylene‑glycol base‑fluid purity, and compatibility with wetted materials. The same review also examined Castrol’s business and global supply‑chain resilience, giving operators confidence in long‑term availability.

“As AI infrastructure scales, the industry needs cooling solutions that are effective, practical to deploy and reliable over the long term,” said Peter Huang, global president of thermal management & data centres at Castrol.

“We evaluate every thermal management fluid under rigorous test conditions before it reaches the market. The validation from NVIDIA confirms that PG25 (OAT) and PG25T (IAT) hold up to that standard, even at the scale and intensity of today’s AI infrastructure,” added Sung A. Kim, data‑centre global technology manager at Castrol.

Direct‑to‑chip cooling works by circulating fluid through cold plates attached to components such as CPUs and GPUs. Castrol describes the fluid as a closed‑loop system that circulates continuously without atmospheric exposure, minimizing evaporative water loss and improving the water‑usage effectiveness (WUE) of AI factory infrastructure.

How the two approaches fit into today’s AI‑compute ecosystem

The Trane‑Eaton reference design tackles the infrastructure layer – power distribution, chillers, and control logic – while Castrol’s fluids address the component‑level heat‑removal path. Together they illustrate a broader industry shift toward tightly coupled electrical‑thermal solutions that can be replicated across multiple sites.

By improving energy efficiency, reducing material consumption, and offering validated liquid‑cooling media, both initiatives give data‑center builders concrete levers to curb capital and operating expenditures as compute density climbs.

Conclusion

The announcements from Trane/Eaton and Castrol add measurable, vendor‑backed data points to the AI‑compute cooling conversation. With quantified efficiency gains, material savings, and a clear validation path for liquid‑cooling fluids, operators now have a more predictable toolkit for scaling high‑performance AI workloads.

Sources

This article was researched and fact-checked against the following sources: