Mitsubishi Heavy Industries (MHI) is deploying advanced cooling technology designed specifically for AI data centers, marking a major shift in how the industry tackles one of its most pressing infrastructure challenges. The company shipped a 10-megawatt-class centrifugal chiller test unit to the United States, with arrival scheduled at the Port of Brunswick, Georgia around July, as part of a broader strategy to support high-density AI workloads that conventional cooling systems cannot handle.

The chiller represents more than just a piece of equipment; it signals a fundamental rethinking of how industrial-scale AI infrastructure must be engineered. As AI factories accelerate deployment globally, the power and cooling demands have grown so extreme that traditional data center approaches are becoming obsolete. MHI is positioning itself as a critical partner in what NVIDIA calls "AI factory-scale" infrastructure, where compute, power, and cooling must function as an integrated system rather than separate components.

Why Are Data Centers Struggling With Cooling?

The challenge is straightforward but severe: AI training and inference require enormous amounts of computing power concentrated in small physical spaces, which generates intense heat. Unlike traditional data centers that might spread workloads across multiple facilities, AI factories pack thousands of graphics processing units (GPUs) into dense clusters, pushing cooling systems to their limits. Water-based cooling has become essential, but it introduces new complications around water scarcity, regulatory approval, and operational complexity.

In Indiana, where Amazon Web Services is building Project Rainier, a 1,200-acre AI data center cluster, the facility is expected to consume up to 2.2 gigawatts of energy annually, enough to power approximately 1 million homes. That scale of power consumption generates proportional cooling demands, and communities are increasingly concerned about water usage and environmental impact. Companies typically keep water consumption data confidential, citing trade secrets, which has made it difficult for regulators and residents to assess the true environmental footprint.

What Makes MHI's Modular Chiller Plant Different?

MHI's solution centers on its Modular Chiller Plant (MCP), a pre-engineered cooling system that integrates the centrifugal chiller with pumps, heat exchangers, and controls into a single modular architecture. The design is built to align with NVIDIA's DSX (NVIDIA's AI factory-scale platform) cooling standards and supports liquid-cooled AI deployments while simplifying installation and enabling scalable expansion.

The MCP architecture offers several practical advantages for hyperscalers and colocation operators:

• Free-Cooling Operation: The system can leverage ambient conditions to reduce active cooling requirements, improving Power Usage Effectiveness (PUE), a standard metric measuring how much energy a data center uses relative to the computing power it delivers.
• Water Efficiency: The closed-loop configuration addresses growing concerns around water scarcity and Water Usage Effectiveness (WUE), reducing the facility's dependence on external water supplies and minimizing environmental impact.
• Reduced Integration Burden: By arriving as a pre-assembled modular unit, the MCP reduces the complexity and time required for on-site installation, allowing operators to focus resources on compute and power infrastructure rather than custom cooling engineering.

The centrifugal chiller itself is based on proven technology with a strong track record in mission-critical applications, according to MHI. The company is currently pursuing U.S. safety and regulatory certification, including Underwriters Laboratories (UL) certification, with the shipment of the test unit contributing to certification readiness ahead of commercial deployment.

How Is NVIDIA Shaping the AI Infrastructure Ecosystem?

MHI's partnership with NVIDIA reflects a broader industry recognition that AI infrastructure requires ecosystem-level coordination. NVIDIA's DSX platform unifies design, simulation, operations, and ecosystem technologies, creating a standardized framework that cooling, power, and compute vendors can build around. MHI participates in the NVIDIA Partner Network as a Power and Cooling Partner, working alongside other ecosystem members to advance integrated solutions.

"AI factories require compute, power, and cooling to be designed as one system. NVIDIA DSX provides the platform for that system-level approach. MHI's work in large-scale cooling and 800 VDC power infrastructure can help ecosystem partners build more scalable and energy-efficient AI factories," said Vladimir Troy, Vice President of AI Infrastructure at NVIDIA.

Vladimir Troy, Vice President of AI Infrastructure at NVIDIA

Beyond cooling, MHI is also advancing 800-volt direct current (800VDC) power architecture, built on decades of experience in large-scale power infrastructure and transportation systems. This higher-voltage approach reduces transmission losses and improves efficiency compared to conventional power distribution in data centers.

What Are the Broader Implications for AI Infrastructure?

The cooling challenge is not merely technical; it has become a bottleneck for AI expansion. Communities hosting data centers are raising legitimate concerns about power consumption, water usage, and infrastructure costs. In Indiana, Governor Mike Braun has stated that he wants the state to lead in the "AI race" without raising electricity bills for residents, expecting companies like Amazon to "pay for every cent of their new power". However, critics note that Indiana currently lacks legislation to enforce such protections, and there are concerns that infrastructure costs could eventually shift to consumers if a facility closes early.

"Indiana has very few safeguards in place when it comes to protecting consumers and protecting our environment, so we would need considerably beefed up regulations and policy to protect folks from big tech data centers before we can be confident that they wouldn't have a very large negative impact on folks," said Bryce Gustafson, an organizer for Citizens Action Coalition.

Bryce Gustafson, Organizer for Citizens Action Coalition

MHI's cooling innovations address the technical side of this equation, but the policy and community impact questions remain unresolved. As data center deployments accelerate, the availability of proven cooling solutions like the MCP may help reduce some operational risks and environmental concerns, but only if paired with stronger regulatory frameworks and transparent water and energy reporting.

The shipment of MHI's 10-megawatt chiller test unit represents a critical milestone in making AI infrastructure more scalable and efficient. However, the real test will come when these systems are deployed at scale across multiple facilities, and communities can assess whether the cooling innovations translate into genuine environmental and economic benefits.