OCP APAC Summit Highlights AI Data Center Management as Power Constraints Grow

The Open Compute Project (OCP) is addressing the challenges of managing AI data centers as they become denser and face power constraints, particularly in the Asia-Pacific region. At the OCP APAC Summit in Taipei on August 11, the focus was on scaling hardware management for AI systems, power, cooling, and server architecture. OCP's Scalable Cloud Infrastructure Management (SCIM) project is tackling issues that arise in facilities with 100,000 nodes or more, including how to manage hardware data across large networks and the impact of technologies like AI, machine learning, RDMA, PCIe/CXL, and DDR on fleet operations.

Regional growth in data centers is significant, with Asia-Pacific increasing its capacity by 7,103 MW in the first half of 2026 to reach 26,455 MW. However, vacancy rates remain high, falling from 10.9% to 10.3%. Despite this growth, power availability remains a critical constraint. An example from GEICO, an insurance company, shows how deploying OCP hardware can lead to substantial cost savings in compute and storage, but also requires new infrastructure management processes and skills.

Power constraints are influencing where new data centers can be built, with some regions facing energy limitations. For instance, Singapore may face energy constraints on additional data center development through 2028. As AI systems become more dense, rack densities for AI workloads could reach around 135 kW, compared to 15–30 kW for typical cloud deployments.

This shift is driving the adoption of liquid cooling to manage higher heat loads. Deloitte recommends shifting workloads based on energy conditions and incorporating flexibility in design to accommodate these changes.