Across regions, governments and tech ecosystems are grappling with how to scale AI compute in practice, exposing gaps between capacity and actual utilization. In Kubernetes-based training at scale, network placement and cross-zone traffic can waste GPUs despite healthy pod signals, highlighting how infrastructural choices quietly erode efficiency. Korea’s drive to automate ports with AI and a national push for onboard autonomous driving chips faces supply frictions, including Nvidia’s Drive Thor shortages that threaten program timelines. At the policy level, Korea also announced a Physical AI Port Strategy and a dedicated Advanced Port Technology Research Institute to operationalize AI across port operations. In India, the ambitious IndiaAI Mission has accrued tens of thousands of GPUs, yet providers report slow user uptake and bureaucratic barriers, with underused capacity and mismatches between subsidized access and early-stage needs. The overall trend is that large compute deployments exist, but mechanics—supply channels, access policies, and cross-domain integration—remain critical bottlenecks to turning capacity into active, impactful AI applications.