The International Energy Agency (IEA) recently released the report “Powering Artificial Intelligence: Strategic Pathways for Data Centers in Power Systems,” covering 34 policy jurisdictions globally and approximately 90% of worldwide data center capacity. The report reviews policy practices and development trends in data center grid connections, energy efficiency management, and cost allocation.
The report indicates that the AI boom is driving continuous growth in data center investment. In 2026, capital expenditure by global hyperscale data center operators is expected to exceed 700 billion US dollars, with rapidly growing power demand posing comprehensive challenges to grid planning, operation, and regulatory systems across countries.
A “queue swelling” phenomenon in data center grid connection applications is common across major global computing markets, with cumulative capacity of pending connection projects in multiple regions already exceeding existing peak system capacity. Many connection applications carry speculative attributes: developers often lock in multiple connection slots in advance to secure resources, but some projects are not bound to specific customers or computing demand. AI computing technology pathways, commercialization progress, and profitability prospects remain uncertain, further amplifying the difficulty of load forecasting. The geographic concentration of data centers further intensifies local grid pressure — in 2025, approximately 60% of global data center capacity was concentrated in industrial clusters; if all construction projects and fully planned projects come online, this share will rise to 70%. A large volume of high-load projects located in areas outside traditional load centers is profoundly changing the topology of grid load structures.
In the traditional model, data centers have been passive electricity consumers. However, under tightening grid constraints, policies and markets are redefining the relationship between data centers and the grid. Non-firm grid connections represent the most promising grid expansion pathway, allowing system operators to curtail data center load under predefined trigger conditions. The IEA estimates that by allowing systems to limit data center usage during power shortage periods, an additional 750 to 900 gigawatts of grid capacity could be released globally. Norway, South Australia, Spain, and other regions have already piloted various modes of non-firm connections. The “on-site power generation” model is also evolving from corporate strategy to policy requirement: Ireland mandates that some new data centers must build or contract dispatchable power sources to ensure system adequacy; China and Spain have set mandatory proportions for renewable energy procurement in new data centers.
However, the promotion of computing-side flexibility still faces significant economic barriers. The annual lease cost for AI-optimized data center capacity can reach 200 billion to 300 billion US dollars per gigawatt, 100 times the cost of storage batteries of equivalent capacity. Globally, there are now 134 regulations governing data center integration with power systems, with nearly 40% released or implemented within the past 12 months. Energy and water efficiency are the earliest regulatory focus areas, with countries generally using power usage effectiveness (PUE) and water usage effectiveness (WUE) as core metrics. China requires newly built data centers at national hub nodes to maintain a PUE of no higher than 1.2. Cost allocation remains the core of current policy debates, as data center connections often require new substations, lines, or grid strengthening projects — who bears the cost directly affects residential electricity prices. Twenty-four US states have approved or are studying large-load-specific electricity rates, based on the principle of “traceable costs” to lock grid investment risk within the connecting enterprise itself.
Source:China Energy News Network