Operating at the Frontier: Scaling the AI Data Center 

  • Scaling AI data centers requires clear organizational roles, centralized operations, and consistent standards across a growing portfolio of sites.
     

  • Better diagnostics, telemetry, and AI-assisted maintenance tools can help operators identify GPU and infrastructure issues faster and improve day-to-day operations. 
     

  • Standardized, modular infrastructure and factory-style deployment can accelerate capacity delivery, improve consistency, and make handoffs between engineering and operations smoother. 
     

  • Large campuses require simpler, repeatable processes and new approaches to hiring and training local talent for specialized operational roles. 
     

  • Operators need to decide which capabilities to manage internally while ensuring external providers can support growth toward gigawatt-scale campuses. 


Tim Hall, VP, Data Center Operations, Meta and Edward Hamner, North American Digital Infrastructure Programs Manager, International Data Center Authority (IDCA)

Tim Hall, VP, Data Center Operations, Meta and Edward Hamner, North American Digital Infrastructure Programs Manager, International Data Center Authority (IDCA)

Reality Check: Evolving the Grid to Meet Growing Power Demand 

Shaneez Mohinani, VP, Head of Strategy & Operations, GridCARE, Tyler Norris, Head of Market Innovation, Google, Chelle Izzi, CCO, PG&E, Bobby Hollis, Chief Development Officer, STACK Infrastructure and Neil Millar, VP, Transmission Planning & Infrastructure Development, California ISO Shaneez Mohinani, VP, Head of Strategy & Operations, GridCARE; Tyler Norris, Head of Market Innovation, Google; Chelle Izzi, CCO, PG&E; Bobby Hollis, Chief Development Officer, STACK Infrastructure and Neil Millar, VP, Transmission Planning & Infrastructure Development, California ISO   

  • Data center power demand is growing faster than traditional grid planning processes can accommodate, putting pressure on utilities to accelerate interconnections. 
     

  • More transparent, real-time operational data can help utilities plan capacity more accurately while protecting reliability as data center loads grow. 
     

  • Software and better utilization of existing grid assets could unlock significant additional capacity without relying entirely on new transmission infrastructure. 
     

  • Flexible data centers can improve time-to-power by shifting compute, using storage, or adding generation when the grid is under stress. 
     

  • Utilities can connect projects faster by directing data centers toward locations with available grid capacity and making greater use of existing infrastructure. 

Power and Cooling at Scale: Supporting the Next Wave of AI Infrastructure 

  • Rising AI rack densities are pushing traditional cooling to its limits, making liquid cooling increasingly important for next-generation infrastructure. 
     

  • Air and liquid cooling will coexist, but the most advanced AI platforms increasingly require component-level liquid cooling to manage different power and temperature profiles. 
     

  • Power and cooling systems need to balance efficiency with capacity, location, connectivity, and the requirements of future AI hardware. 
     

  • Chip-level telemetry can help operators match cooling infrastructure more closely to actual demand, reducing unnecessary overprovisioning while maintaining reliability. 
     

  • Standardized liquid-cooling interfaces and closer integration of compute, power, and cooling can simplify deployment and improve efficiency across the entire system. 


Jeff Moore, VP, Strategic Partnerships, Aegis Cooling, an Airsys Company, Ryan Brown, Director, Product Management, Phononic, Steve Madara, Distinguished Engineer & VP Thermal, Vertiv, Phil Read, Senior Director, Colocation Product Management, Equinix and Racha Lwali, Head of Data Center Liquid Cooling, AWS

Jeff Moore, VP, Strategic Partnerships, Aegis Cooling, an Airsys Company; Ryan Brown, Director, Product Management, Phononic; Steve Madara, Distinguished Engineer & VP Thermal; Vertiv, Phil Read, Senior Director, Colocation Product Management; Equinix and Racha Lwali, Head of Data Center Liquid Cooling, AWS 

Enterprise AI Capacity Planning: Procurement, Platform Choices, and Operational Reality 

 

Naveed Saeed, VP, Global Technical Management, Uptime Institute, Jake Ring, Co-Founder, President & CEO, Power and Data Management, Mike Bahr, Head of Infrastructure, Data Center & Networking, DRW, Sonia Hingorany, GM | Partner, Azure Cloud Operations + Innovation, Microsoft and Dave Driggers, CEO, Cirrascale

Naveed Saeed, VP, Global Technical Management, Uptime Institute; Jake Ring, Co-Founder, President & CEO, Power and Data Management; Mike Bahr, Head of Infrastructure, Data Center & Networking, DRW; Sonia Hingorany, GM | Partner, Azure Cloud Operations + Innovation, Microsoft and Dave Driggers, CEO, Cirrascale    

  • Enterprise AI capacity is increasingly constrained by access to power, permitting, grid connections, equipment availability, and skilled labor. 
     

  • Infrastructure planning should focus on useful AI work delivered per dollar and unit of power, rather than optimizing individual hardware or facility metrics. 
     

  • Enterprises need to plan workloads, infrastructure, operations, hardware refresh cycles, and budgets together while ensuring suppliers can support smaller deployments as well as hyperscale projects. 
     

  • Existing data centers can support more AI workloads through upgrades such as liquid cooling and improved power distribution, helping unlock capacity without building entirely new facilities.
     

  • Different workloads require different infrastructure: inference can often run in existing medium-density facilities, while large-scale training typically requires specialized, high-density clusters.