Powering AI at scale – practical strategies for today’s data center energy challenges

Jeff DreesCEO, Mission Critical GroupLinkedIn

Didiar PadillaRegional Director, iMasonsLinkedIn

As AI workloads grow, so does the strain they place on data center power systems. Many organizations are discovering that the real barrier to scaling isn’t compute capacity, it’s the ability to deliver reliable, cost-effective and sustainable power at the pace AI demands. Grid queues are getting longer, energy costs are unpredictable, and high-density clusters are pushing traditional power and cooling designs to their limits.

This roundtable brings together key stakeholders to discuss the power challenges facing modern AI data centers and explores what operators can do about them right now: grid constraints, electrical infrastructure pressure, and the operational risks of running increasingly power-hungry environments, alongside practical approaches like on-site generation, microgrids, new distribution architectures, and smarter energy management.

• Power-density characteristics of modern AI clusters – transient load behaviours and implications for upstream electrical design
• Grid connection constraints, transformer lead times, substation design
• Trade-offs between traditional architectures and direct-to-chip power delivery
• Integration of on-site generation
• Liquid cooling options
• Approaches to energy storage integration
• Monitoring and control strategies for maintaining stability in high-density, fluctuating AI environments

• Evaluate the power profile of AI-heavy environments
• Plan for growth when grid capacity is limited
• Weigh the trade-offs of alternative energy sources and on-site solutions
• Build resilient, efficient power systems that can support high-density compute
• Align energy strategy with sustainability and business goals

Delivering AI at scale – navigating the strategic shift toward edge-based inferencing

Rodney DellingerCTO & Head of Architecture, Web, NokiaLinkedIn

Scott WillisCEO, DartPointsLinkedIn

As AI adoption accelerates, the industry is seeing a shift from large, centralized training clusters to distributed inferencing that runs closer to users, devices, and real-time operations. This move toward edge-based AI is driven by the need for lower latency, reduced backhaul traffic, and faster decision-making, but it also introduces new architectural, operational, and investment considerations for organizations building modern AI infrastructure.

This roundtable examines how inferencing workloads are evolving, focusing on the interplay between compute, connectivity, and placement. We will also examine where inferencing should run, how edge sites fit into the broader AI ecosystem, and what infrastructure is required for consistent performance across distributed environments.

• How inferencing differs from training in terms of workload behavior, latency sensitivity, and deployment models
• The growing role of regional edge sites, micro-edge locations, and distributed compute fabrics
• Why fiber availability, optical capacity, and diverse routing are becoming critical differentiators for AI-ready networks
• Network and connectivity requirements for real-time inferencing, including bandwidth, latency, and traffic-flow considerations
• How to decide which workloads belong at the edge versus centralized data centers
• Operational challenges such as lifecycle management, model updates, telemetry, and security across distributed nodes
• Infrastructure strategies that balance performance, cost, and scalability as inferencing volumes grow
• Partnership models with hyperscalers, data-center operators, and enterprises to accelerate AI deployment at the edge
• The role of telecom operators – fiber networks as premium infrastructure for AI workloads requiring deterministic performance