TeraWulf Turns Former Coal Plant Site Into 500 MW AI Campus With Liquid Cooling

From available megawatts to usable AI capacity

Andrew Whitmore, Vice President of Sales at Motivair by Schneider Electric, argues in a company blog post that available megawatts do not automatically become AI data center capacity. In his words, the competitive advantage lies in how quickly operators can operationalize that power, which means coordinating electrical, cooling and IT infrastructure.

TeraWulf has expanded its Lake Mariner site on the shores of Lake Ontario near Buffalo from a proof of concept into a large-scale, phased campus. According to the post, such builds typically take around 36 months, but TeraWulf worked with Motivair by Schneider Electric and other vendors to use a scalable liquid-cooling infrastructure that kept pace with an accelerated schedule.

The facility, known as Lake Mariner Data, will deliver 500 MW of gross capacity across six buildings. Plans call for expansion to 750 MW.

Power access gave a head start, design kept the pace

The post says access to power has become one of the biggest constraints on new AI data center development as demand and rack densities rise. It quotes TeraWulf as saying it prioritizes sites with access to utility-scale power and transmission infrastructure, which "can materially shorten development timelines relative to traditional greenfield projects."

Power alone was not enough, according to the author. Rapid deployment also depended on choosing the right cooling architecture and coordinating mechanical, electrical and IT systems early. The aim was to support higher rack densities, future GPU generations, phased expansion and long-term efficiency. With a compressed timeline, any deviation could delay the project's critical path.

The cooling platform at Lake Mariner

Motivair says it brought decades of experience in high-density supercomputing environments, combined with proven rear door heat exchanger (RDHx) technology. The company argues this gave TeraWulf confidence that the thermal infrastructure could keep pace with the build, perform reliably and scale as compute requirements change.

Motivair ChilledDoor technology is a key part of the platform and delivers up to 75 kW of cooling directly at the rack. In three of the Lake Mariner buildings, the system uses fans to draw room air through the servers while cool facility water circulates through the rear-door heat exchanger. According to the post, this achieves 100% heat removal and room-neutral cooling.

• In-rack manifolds
• In-rack coolant distribution units (CDUs)
• A custom water-cooled loop for direct liquid-cooled (DLC) servers

The post says the goal was an architecture that could be repeated and adapted as more AI capacity comes online, not only a cooling system for the first phase.

What it means for future AI data centers

Motivair draws a broader lesson: thermal infrastructure is a strategic architectural decision that influences deployment speed, scalability and operational flexibility, and should be part of planning from Day One. The company notes that AI compute evolves faster than the facilities that house it, and each hardware generation raises rack densities and thermal demands.

For operators racing to build AI capacity, the post's message is that electrical, mechanical, cooling and IT systems need to be planned together so they can scale together. The account comes from a vendor involved in the project, so its claims about timelines and performance reflect Motivair's own view.