How Trane and Eaton Are Integrating Power and Cooling for AI Data Centers

AI data centers are increasingly constrained by two linked infrastructure problems: delivering large amounts of power to dense computing equipment and removing the heat that power generates. Trane Technologies and Eaton’s new reference design addresses both as a single “grid-to-chip” system rather than as separate electrical and cooling projects.

The companies describe the architecture as an industry-first collaboration for AI factories built around NVIDIA’s DSX AI Factory Reference Design. It is important to distinguish that announcement from a customer deployment: the available information describes a reference architecture incorporated into the companies’ platforms, not a disclosed installation at a named data-center site.

What Trane and Eaton announced

The collaboration combines Eaton’s electrical-system architecture with Trane’s thermal-management design. Its central change is a coordinated medium-voltage approach intended to replace more traditional, siloed low-voltage planning for high-density AI facilities. Power and cooling systems share data and controls so they can respond to operating conditions together rather than being designed and operated independently.

The design is aligned with the NVIDIA DSX AI Factory Reference Design and the NVIDIA Omniverse DSX Blueprint. That gives the companies a common digital and engineering framework for planning the electrical, thermal and control infrastructure needed by next-generation AI facilities.

How the medium-voltage architecture works

In a conventional data-center design, electrical distribution and cooling are often treated as separate engineering workstreams. That separation can make it harder to match cooling capacity with actual computing loads, coordinate equipment changes or optimize the facility as a whole.

Trane and Eaton’s approach links the two systems earlier in the design process:

• Eaton provides the power-distribution architecture, using medium-voltage distribution to support high-density loads and reduce reliance on more copper-intensive layouts.
• Trane provides the thermal-management architecture, including designs intended for AI factories and liquid-cooling configurations.
• Shared controls and data coordinate operation, allowing power and cooling equipment to respond to changing facility conditions in real time.
• The design anticipates newer power and cooling configurations, including liquid cooling and emerging direct-current architectures.

The practical idea is to optimize the path from the utility grid to the computing chip as one system. That can help designers account for electrical losses, heat production, cooling capacity, cable requirements and equipment placement before construction begins.

The claimed savings: 15%, 30% and 80%

Compared with conventional low-voltage designs, Trane and Eaton say the reference architecture can deliver up to:

• 15% better energy efficiency;
• 30% lower installation costs; and
• 80% less copper use.

Those figures should be read as company-reported design estimates, not as independently validated performance measurements from an identified operating data center. Actual results would depend on factors such as facility scale, electrical topology, cooling technology, local conditions and how closely a deployment follows the reference design.

Even with that qualification, the targets show where the companies see the main economic opportunity. Lower copper use can reduce material volume and cabling requirements. A more coordinated design may also reduce duplicated engineering, construction complexity and mismatches between installed power and cooling capacity.

Where the design is available

The joint architecture is incorporated into two platform offerings:

• Trane Continuum Rubin DSX provides Trane’s thermal reference designs for AI factories.
• Eaton Beam Rubin DSX is Eaton’s pre-engineered platform for converging power, cooling and controls, with the goal of bringing AI-factory capacity online faster.

Eaton’s distribution technology also supports the Trane implementation, linking the two companies’ contributions within the broader DSX-aligned design approach.

This does not mean every customer deployment will use an identical equipment configuration. A reference design is a repeatable engineering template, not a guarantee that every project will be built in exactly the same way.

How Omniverse fits into the design process

The electrical-and-thermal collaboration builds on Trane’s earlier work with NVIDIA Omniverse. Trane said it improved the efficiency of its AI-factory thermal reference design by nearly 10% using Omniverse-based design and simulation, while also introducing two additional Continuum Rubin DSX reference designs.

The newer collaboration extends that digital-design effort beyond cooling. Instead of simulating thermal performance in isolation, the reference architecture is intended to connect electrical distribution, thermal management and controls in a more unified model.

That matters because AI-factory performance is increasingly determined by interactions between systems. A change in computing density can affect power demand; power demand affects heat output; heat output affects liquid-cooling capacity, pumping requirements and facility efficiency.

What Eaton’s Boyd Thermal acquisition changes

Eaton completed its $9.5 billion acquisition of Boyd Thermal on March 12, 2026. The deal added liquid-cooling capabilities to Eaton’s electrical and power-management portfolio and strengthened its position as an end-to-end data-center infrastructure provider.

That broader portfolio is strategically relevant as AI systems move toward higher power density and more direct liquid-cooling requirements. It gives Eaton a stronger basis for connecting medium- and high-voltage infrastructure with thermal technologies across the data-center stack.

However, the acquisition should not be interpreted as evidence that every Trane–Eaton reference-design deployment will use Boyd equipment. The joint announcement establishes an integrated design approach; it does not identify a specific customer project or prescribe one universal bill of materials.

Why this matters as AI factories scale

The infrastructure challenge is not simply finding more electricity or installing more cooling. Operators must deliver power efficiently, distribute it across increasingly dense computing loads and remove heat without inflating construction and operating costs.

That is why the power-and-cooling connection is becoming a central design issue. A coordinated architecture can help operators evaluate:

• how much power reaches the computing equipment;
• where conversion and distribution losses occur;
• how much heat each computing zone produces;
• whether liquid cooling is required; and
• how much electrical and thermal capacity should be installed for future expansion.

The Trane–Eaton design does not eliminate those decisions. Its significance is that it packages them into a shared reference framework, potentially making it easier to compare facility designs before construction and adapt them as AI hardware evolves.

The bottom line

Trane and Eaton are proposing that AI data centers should be designed from the grid to the chip as one coordinated electrical and thermal system. Their NVIDIA DSX-aligned reference design combines Eaton’s medium-voltage distribution with Trane’s thermal-management architecture, shared controls, liquid-cooling support and compatibility with emerging direct-current systems.

The headline targets—15% better energy efficiency, 30% lower installation costs and 80% less copper—are substantial, but they remain vendor claims tied to a reference design rather than independently verified results from a disclosed facility. The more durable takeaway is the design shift itself: as AI power density rises, power delivery and heat removal increasingly have to be engineered together.