AI is quickly becoming one of the biggest strains on the U.S. power grid and health systems are not immune to the fallout. As hospitals move AI from pilot programs to enterprise-wide deployment, rising demand, grid constraints and yearslong interconnection timelines mean the infrastructure decisions made today could determine how reliably they can scale AI five to ten years from now.
Becker’s Healthcare spoke with Greg Castle, U.S. power & grid commercial leader at Schneider Electric, and James Vun Cannon, healthcare segment manager for strategic accounts and U.S. digital power at Schneider Electric, about what’s driving the surge in AI-related power demand and why facilities, IT, clinical and finance leaders need to plan infrastructure and energy together.
Editor’s note: Responses have been edited for length and clarity.
Question: What’s driving the current spike in AI-related power demand, and where is it showing up most acutely?
Greg Castle: In the last three years, 487 gigawatts of data center projects have been announced for the next five years, even half of that is remarkable. About 150 gigawatts is planning to go off-grid entirely because of long interconnection queues, which is roughly the equivalent of France and the UK combined, or 25% of total U.S. peak demand. T
The ripple effects will reach every industry, healthcare included.
Q: What’s the biggest misconception you see as health systems try to scale AI?
James Vun Cannon: That AI readiness is primarily an IT initiative. In reality, it very quickly becomes a facility, infrastructure and operations initiative.
Power capacity, cooling and resiliency planning all have to scale with it. Health systems need long-term capacity planning now, not after the next study. They need a 1-3-5 plan in place.
Q: How should hospitals think about on-premise versus cloud for AI workloads?
JVC: Hospitals are required by code to defend in place, so they can’t put every decision in the cloud and risk losing communications during an outage. Most will land on a hybrid model, weighing latency, data governance, cybersecurity and clinical response time, some combination of cloud, edge and on-prem rather than one approach.
Q: What does this mean in practical terms for a hospital’s energy needs?
GC: A typical 500-bed hospital runs 8 to 10 megawatts today. Fully deploying AI could add a 1-to-2-megawatt on-prem data center for low-latency needs, a 10% to 25% increase in energy for AI alone. Utilities may not have the luxury of granting extra megawatts on the timeline this industry is moving at, so looking 5 to 10 years ahead matters.
Q: What’s a realistic starting point for a leader who wants to get ahead of this?
JVC: AI readiness and energy readiness are becoming inseparable. Healthcare has always focused on resiliency for patient care and nothing in a hospital runs without electricity. The organizations that plan for both now will be the ones positioned to scale AI safely and reliably.
Facilities and IT have historically worked in silos. AI is forcing new alliances between them, plus cybersecurity, clinical leadership and finance, building the investment plan with the CFO early, before capacity runs out.
GC: Microgrids are moving from nice-to-have to critical, and options like energy-as-a-service can de-risk energy without heavy capital outlay, while also protecting against rate increases.
Learn more: https://www.se.com/us/en/work/solutions/healthcare/
