AI data centers are no longer assembled; they’re designed as integrated systems. Using simulation, digital twins, and GPU-accelerated modeling, operators can now model how power, cooling, and compute interact before a single piece of equipment is deployed. At the same time, the industry is entering what NVIDIA has called an “inference inflection point” as workloads move from training into large-scale, real-time deployment across industries—from industrial automation and robotics to geospatial intelligence and foundational digital infrastructure. Together, these shifts are redefining what AI infrastructure must deliver—and how precisely it must be designed in advance. But the transition from design to deployment remains the industry’s hardest problem.

This fireside brings together leaders across hyperscale infrastructure, AI systems design, and open hardware ecosystems to examine:

• How digital twins and simulation are shaping next-generation AI facilities

• How Meta and others are deploying AI infrastructure at global hyperscale

• The role of OCP and open systems collaboration in accelerating deployment

• Where designs break when they hit real-world construction, operations, and grid constraints

Framed through “designed vs. deployed,” the session looks at the widening gap—and emerging convergence—between what can be precisely modeled, and what can be reliably delivered at hundreds of megawatts.