Beyond the Data Centre: How AI Is Reshaping Digital Infrastructure Digital Infrastructure trends for the future From Coal Power to AI Infrastructure Scale, Capacity and Grid Capacity Why AI Is Changing Infrastructure Design AI Infrastructure and the Public Debate Billy Linehan More about Irish Tech News

A visit to TeraWulf's Lake Mariner campus reveals how AI infrastructure is evolving far beyond the traditional data centre

Standing beside the former Somerset coal-fired power station on the shores of Lake Ontario, it was difficult to miss the scale of what is now taking shape. (See photo above of the Lake Mariner facility under construction).

Construction crews were working across the Lake Mariner campus, where TeraWulf is transforming a site once associated with coal-fired electricity generation, and later bitcoin mining, into a major AI and high-performance computing facility.

One of the largest buildings currently under development, known as CB-4, spans approximately 330,000 square feet, equivalent to more than four full-sized football pitches under one roof.

The site was one of several stops during a Schneider Electric-hosted visit to Buffalo, New York examining the infrastructure emerging around large-scale AI computing.

While much public discussion focuses on AI software and increasingly powerful processors, the visit highlighted something less visible: the industrial infrastructure now being built to support the next generation of AI systems.

Lake Mariner's story is also one of industrial regeneration. Rather than developing a completely new location, TeraWulf is repurposing an established industrial site, reusing land, transmission infrastructure and grid connections already associated with power generation.

The company has stated that its operations are powered predominantly by zero-carbon electricity, drawing on hydroelectric and nuclear generation available through New York State's electricity system, alongside solar generation currently under development on site.

During a tour of the campus, Sean Farrell, COO of TeraWulf, described a project being delivered at remarkable speed. Around 1,600 people are involved across engineering, construction and specialist trades. "We work around the clock," Farrell said.

According to Farrell, facilities that once took years to deliver are now being brought online in less than 12 months. The speed of development was one of the most striking aspects of the visit.

Walking around the site, it became clear that this was no longer simply a story about data centres. Alongside the buildings themselves were substations, transformers, cooling systems, power distribution equipment and extensive electrical infrastructure.

Having toured Lake Mariner, I asked Robert Bunger, Global Director of Data Centre Solution Architecture at Schneider Electric, whether the industry had reached a point where access to power, cooling and grid capacity now matters as much as constructing new facilities.

Bunger's answer was immediate. "Scale, capacity and grid capacity," he said. "Absolutely."

The response reflected much of what visitors had seen throughout the day. The challenge facing operators is no longer simply creating more data centre space. It is securing enough power, cooling and supporting infrastructure to keep pace with rapidly growing AI workloads.

Speed is also becoming a critical factor. Throughout the visit, Farrell, and later Bunger, returned to the challenge of delivering infrastructure quickly enough to meet demand. The issue is no longer limited to buildings. Power equipment, cooling systems, specialist engineering expertise and supply chain capacity all have to be available at the right time.

Facilities that once took years to deliver are now expected in months. For operators competing to support AI customers, the ability to deploy infrastructure rapidly is becoming a competitive advantage in its own right.

One of the questions I put to Bunger concerned the growing industry discussion around 800 VDC and new high-density power architectures. Was this simply another technical trend, or evidence that traditional data centre electrical architectures were no longer sufficient for AI-scale workloads?

Bunger's answer suggested the latter. "The need to change the way we're doing