Intro

As the data center industry enters the AI era in earnest, incremental upgrades are no longer enough. That was the central message of the Data Center Frontier Trends Summit 2025 session “AI Is the New Normal: Building the AI Factory for Power, Profit, and Scale,” where operators and infrastructure leaders made the case that AI is no longer a specialty workload; it is redefining the data center itself.

Panelists described the AI factory as a new infrastructure archetype: purpose-built, power-intensive, liquid-cooled, and designed for constant change. Rack densities that once hovered in the low teens have now surged past 50 kilowatts and, in some cases, toward megawatt-scale configurations. Facilities designed for yesterday’s assumptions simply cannot keep up.

Ken Patchett of Lambda framed AI factories as inherently multi-density environments, capable of supporting everything from traditional enterprise racks to extreme GPU deployments within the same campus. These facilities are not replacements for conventional data centers, he noted, but essential additions; and they must be designed for rapid iteration as chip architectures evolve every few months.

Wes Cummins of Applied Digital extended the conversation to campus scale and geography. AI demand is pushing developers toward tertiary markets where power is abundant but historically underutilized. Training and inference workloads now require hundreds of megawatts at single sites, delivered in timelines that have shrunk from years to little more than a year. Cost efficiency, ultra-low PUE, and flexible shells are becoming decisive competitive advantages.

Liquid cooling emerged as a foundational requirement rather than an optimization. Patrick Pedroso of Equus Compute Solutions compared the shift to the automotive industry’s move away from air-cooled engines. From rear-door heat exchangers to direct-to-chip and immersion systems, cooling strategies must now accommodate fluctuating AI workloads while enabling energy recovery—even at the edge.

For Kenneth Moreano of Scott Data Center, the AI factory is as much a service model as a physical asset. By abstracting infrastructure complexity and controlling the full stack in-house, his company enables enterprise customers to move from AI experimentation to production at scale, without managing the underlying technical detail.

Across the discussion, panelists agreed that the industry’s traditional design and financing playbook is obsolete. AI infrastructure cannot be treated as a 25-year depreciable asset when hardware cycles move in months. Instead, data centers must be built as adaptable, elemental systems: capable of evolving as power, cooling, and compute requirements continue to shift.

The session concluded with one obvious takeaway: AI is not a future state to prepare for. It is already shaping how data centers are built, where they are located, and how they generate value. The AI factory is no longer theoretical—and the industry is racing to build it fast enough.