The race to build AI infrastructure is creating a problem that cannot be solved with faster chips alone: data centers need power and cooling systems designed around increasingly dense computing workloads. Trane Technologies and Eaton are responding with a joint reference architecture that combines thermal management and electrical distribution for next-generation AI factories, aiming to reduce construction complexity while improving energy and resource efficiency.
AI data center design has traditionally been divided among electrical, mechanical, computing and facilities teams. As AI clusters push rack densities higher, that separation is becoming harder to maintain.
Trane Technologies and Eaton are attempting to address the problem with a coordinated reference design that brings power distribution and thermal management into a single architecture for AI factories. The companies say the design can deliver up to 15% combined energy-efficiency gains, reduce copper use by as much as 80%, and cut installation costs by up to 30% compared with conventional low-voltage approaches. Those figures are company claims and will ultimately depend on deployment conditions and system configuration.
The collaboration is built around two platforms: Trane Continuum Rubin DSX and Eaton Beam Rubin DSX. Both are aligned with NVIDIA’s DSX architecture, with Eaton supplying power-distribution technology for the Trane platform. The companies describe the result as a pre-coordinated electrical and thermal system extending from the grid to the computing equipment.
The important change is less about adding another piece of data center equipment than about changing how infrastructure is engineered.
AI factories require significantly more coordination between compute loads, electrical systems and cooling. A sudden change in computing demand can affect both power consumption and heat generation. Designing those systems independently can create additional engineering work and make it harder to optimize the facility as a whole.
The Trane-Eaton approach attempts to make those relationships explicit. The companies say power and cooling systems can exchange leading indicators and respond more dynamically to changes in computing requirements. In practical terms, that could allow infrastructure teams to design around expected AI workloads rather than treating electrical and thermal systems as largely independent layers.
The timing is significant. McKinsey estimates that global data center capacity demand could nearly triple by 2030, with roughly 70% of that demand driven by AI workloads. The consulting firm estimates that meeting worldwide compute demand could require approximately $6.7 trillion in cumulative data center capital expenditure through 2030, including $5.2 trillion for facilities supporting AI workloads.
That investment creates pressure to build faster, but speed introduces another challenge: standardizing designs without making them inflexible.
This is where NVIDIA’s expanding DSX ecosystem becomes important. NVIDIA’s Vera Rubin DSX reference architecture is designed to coordinate compute, networking, storage, power, cooling and facility controls across AI factories. NVIDIA’s Omniverse DSX Blueprint adds a digital-twin layer, allowing engineers to simulate electrical, thermal, network and site behavior before and during deployment.
Trane and Eaton are therefore entering an ecosystem rather than creating an isolated architecture. NVIDIA lists both companies among the industry partners contributing to the broader DSX architecture and blueprint, alongside infrastructure and engineering companies including Schneider Electric, Vertiv, Siemens, Cadence and others.
The competitive implication is noteworthy.
Companies such as Schneider Electric, Vertiv and GE Vernova are also developing infrastructure around the increasingly integrated requirements of AI data centers. NVIDIA’s own DSX documentation now describes reference designs spanning hardware, facilities infrastructure, simulation and operational controls.
Trane’s traditional strength is thermal management, while Eaton is a major provider of electrical power-management technology. Their collaboration effectively combines two infrastructure domains that AI data center operators increasingly need to optimize together.
Medium-voltage architecture is another important element. Moving power closer to the high-density compute environment can reduce some of the material and conversion requirements associated with conventional low-voltage distribution. But the claimed reductions in copper and installation costs should not be interpreted as universal savings: actual economics depend on facility size, electrical topology, utility requirements, equipment selection and local codes.
For enterprise infrastructure teams, the bigger lesson is that AI deployment is increasingly becoming a facilities engineering decision as much as an IT decision.
A company planning an AI cluster cannot evaluate accelerator performance in isolation. Power availability, rack density, cooling technology, electrical redundancy, construction schedules and operating costs can determine whether the underlying infrastructure can support the intended workload.
The same shift is visible in NVIDIA’s digital-twin strategy. Its Omniverse DSX Blueprint is designed to model physical AI factory infrastructure and simulate power and thermal behavior, while its DSX architecture connects IT and operational technology systems.
For data center operators, this creates a potential path toward more repeatable AI infrastructure. Instead of engineering every facility from scratch, validated reference architectures could become starting points for standardized deployments, with local modifications applied where required.
That could matter as much for time-to-market as the efficiency gains themselves.
The AI infrastructure industry is entering a phase where the constraint is no longer simply how many accelerators can be purchased. It is how quickly organizations can turn land, power, cooling, networking and compute into a functioning AI factory.
Trane and Eaton’s collaboration reflects that broader transition. The winning data center architecture may increasingly be the one that treats power, thermal management and computing as a single system—rather than as separate projects connected late in the construction process.
Market Landscape
The AI data center infrastructure market is shifting toward system-level co-design.
McKinsey says data center capacity could approach 219 GW by 2030 under its continued-momentum scenario, compared with 82 GW in 2025, with AI accounting for the majority of incremental demand.
That growth is putting pressure on three areas:
- Power: AI clusters require increasingly large and predictable electrical capacity.
- Cooling: Higher rack densities are making advanced thermal management increasingly important.
- Deployment speed: Operators need repeatable architectures that reduce engineering and commissioning time.
McKinsey’s research specifically argues that power, cooling and IT components increasingly need to be codesigned rather than treated as separate systems, reinforcing the strategic direction behind the Trane-Eaton announcement.
The competitive field includes NVIDIA’s DSX architecture, Schneider Electric’s power-management and simulation technologies, Vertiv’s thermal and power infrastructure, and solutions from other electrical, mechanical and data center specialists.
The differentiation is moving toward the ability to deliver an integrated architecture—not simply an individual cooling unit, switchgear product or accelerator.
Top Insights
- Trane Technologies and Eaton are integrating thermal management and medium-voltage power systems to simplify deployment of high-density NVIDIA AI factories.
- The companies claim up to 15% energy-efficiency gains, 80% less copper and 30% lower installation costs versus conventional low-voltage designs.
- NVIDIA’s DSX ecosystem is becoming a coordination layer connecting AI compute with power, cooling, networking, simulation and facility operations.
- McKinsey expects global data center capacity demand to nearly triple by 2030, making standardized infrastructure increasingly important for hyperscalers and enterprises.
- Enterprise AI teams will increasingly need facilities, energy and infrastructure engineering involved alongside IT teams when planning large-scale AI deployments.
Power Tomorrow’s Intelligence — Build It with TechEdgeAI
