Data Center Migration

Overview

Data center migration, in the AI infrastructure context, refers to the large-scale geographic relocation of hyperscale AI training facilities away from traditional urban data center clusters (which historically co-located with fiber networks and enterprise customers) toward less-populated, lower-cost-energy inland regions. The primary driver is power cost and availability: AI training clusters at the 50,000–100,000 GPU scale consume tens of megawatts to gigawatts of power, making energy price per kWh a dominant cost variable that overwhelms real estate and connectivity cost considerations.

The consequence is explosive growth in WAN traffic: as compute moves further from end users and across geographies, AI-generated traffic traversing long-haul networks increases dramatically. This simultaneously validates the scale-across network tier requirement (connecting distributed clusters) and increases the relative importance of wide-area networking quality for AI workloads.

Meta's planned Hyperion data center — up to 5 GW in planned capacity, roughly the size of Manhattan — was cited at OCP Global Summit 2025 as an example of the emerging scale of AI training infrastructure that cannot fit within traditional urban data center footprints.