Roundtable: Designing for an Uncertain AI Demand Curve

Christopher Gorthy, DPR Construction:  Flexibility, cost, schedule, reliability are all key elements to support a customer's needs and long-term strategy. We have to understand the priorities of each project and each customer we work with and balance the approach for that effort. Flexibility often comes with added upfront cost, but when it’s applied thoughtfully, teams can try to find ways to offset costs in the short term while also providing meaningful long‑term savings and solutions. 

Engaging with technical experts early on can help customers make decisions or perform studies to balance the right priorities. For example: reducing water usage may increase electrical demand; designing a larger building shell may carry initial cost but accelerate future deployments; adding space in medium‑ or low‑voltage pathways today may enable higher densities tomorrow.

All of these questions have trade-offs, which is why we encourage our customers to not get stuck in ridged decision structures, as regions and geographies should influence key design decisions. A baseline standard is useful, but it must have enough flexibility for the team to execute the mission. The “fastest path” isn’t determined by cost per MW or total duration alone, it’s the sum of hundreds of design, engineering, procurement and deployment decisions that must be evaluated together.

As an industry we must be clearer about the actual impacts and steps many of these projects are taking to reduce energy costs and actually strengthen the power grid, invest in the communities for decades to come, improve other infrastructure that supports the local economy that goes beyond just the building.  When we improve transparency around the data and the outcomes, we can counter misconceptions and create a more collaborative public process.

Miranda Gardiner, iMasons Climate Accord:  Our industry is under pressure to deliver capacity quickly while avoiding infrastructure decisions that could become stranded or inefficient over time. A key starting point is bringing sustainability and delivery professionals into planning early and often, so these decisions reflect both near-term deployment needs as well as long-term energy and emissions goals.

Futureproofing increasingly relies on flexible, modular power and electrical architectures such as designing substations, distribution systems, and cooling infrastructure with clear expansion pathways so capacity can scale in phases as demand materializes. Approaches like this can help operators maintain speed-to-market while avoiding overbuilding, improve energy efficiency, and align infrastructure growth with sustainability commitments.

Policy and energy market dynamics are also shaping how projects plan for long-term power needs. We can assist policymakers to focus on the needs of the digital ecosystem by working with our leaders to streamline grid interconnections and accelerate infrastructure development, particularly in regions with abundant hydroelectric, solar, wind, and natural gas resources.

We are already seeing new policies in some markets driving investment in microgrids, behind-the-meter generation, and battery energy storage systems (BESS) that help stabilize the grid while supporting lower-emissions operations. Expanded use of energy storage can also reduce reliance on traditional power supply inside our facilities, freeing up room for more high-density compute.

By combining modular infrastructure design, early sustainability integration, and proactive engagement with policymakers and utilities, the industry can future-proof both AI campuses and communities while advancing reliability, efficiency, and emissions reductions. 

Miles Whitling, Maddox Industrial Transformer:  On the one hand, power infrastructure requires much less flexibility than the design for the data hall itself. While compute technology, rack density, and cooling strategies continue to evolve, power infrastructure remains relatively stable. Once the core power systems are installed, they typically remain consistent over the life of the data center. And given the pace of demand, developers probably don’t need to fear overbuilding power capacity–if it’s available, it will get used.

At the same time, developers still need to align engineering and procurement teams to ensure equipment designs can serve multiple sites. Most hyperscalers are already standardizing their specifications and ordering long-lead equipment like transformers and switchgear in large quantities, assigning gear across multiple projects as needed. This can significantly accelerate delivery.

Mike Connaughton, Leviton Network Solutions:  Future-proofing means designing with enough flexibility to incorporate new technologies when they become available. Achieving this in an efficient, cost-effective manner requires a foundation of infrastructure with a modular design. With cabling infrastructure, modular systems and pre-terminated structured cabling enable incremental scaling and simplified installation, significantly improving the overall speed of deployment.

And it allows data centers to avoid costly rip-and-replace projects and immediately begin leveraging new technologies with minimal downtime. Fixed designs and complex cabling will struggle to keep pace, leading to inefficiencies, compatibility issues and a critical lack of scalability.

NEXT:  The Next Credibility Test