AI’s Power Bill: Singapore’s Grid Planning Faces Surging Data Center Demand | Singapore Power Grid AI Demand

Singapore’s electricity planning is entering a new phase as AI-driven data centres expand and run continuously. One projection cited by Asian Power says Singapore’s electricity demand linked to this AI-driven data centre growth is expected to nearly triple between 2025 and 2030. That scale matters for grid planning because these loads are not only large, but also persistent, and they often cluster in specific areas. The result is that even if demand growth looks manageable in aggregate, the grid can still face sharp, localised spikes in peak demand that strain transmission infrastructure and increase the urgency of reinforcements.

Cost pressure is already part of the story. The same report notes electricity prices increasing by up to 12% as supply pressures build, a reminder that demand surges can be felt quickly by the market. At the system level, electricity demand overall is projected to grow at around 3.2% annually, with supply expected to grow faster. But the operational challenge is timing and flexibility: many grid systems are not yet able to respond quickly enough to fluctuations in demand, which elevates the value of smarter controls and greater automation, especially when data centres can shift the load profile and the daily peaks.

Why AI Loads Create New Grid Planning Stress

Research on AI data centres highlights why grid planners worry about these facilities across multiple timescales, from long-term planning and interconnection to short-term operations and real-time stability. The loads are shaped by different stages of AI work, including training and inference. As context outside Singapore, one review estimates training GPT-3 consumed 1.29 GWh, while training GPT-4 rose to an estimated over 50 GWh, described as nearly 0.1% of New York City’s annual electricity use. These examples illustrate how fast model-scale changes can translate into very different electricity requirements, complicating forecasting and capacity planning.

Global figures reinforce the trend, while also showing why Singapore’s planners are watching international developments. The IEA is cited as estimating global data centres consumed around 415 TWh in 2024, about 1.5% of total global demand, and projecting consumption to more than double by 2030 to around 945 TWh, with AI identified as the primary driver. Separately, Goldman Sachs is cited projecting AI will drive a 160% to 165% increase in data centre power demand by 2030 compared with 2022 levels. Several jurisdictions, including Singapore, have introduced restrictions on new data centre connections given grid capacity constraints.

Against this backdrop, Singapore is already applying AI to grid operations, which links directly to how the Singapore power grid AI demand story evolves. A Scientific Reports study describes an AI system that uses historical weather data, real-time solar irradiance measurements, satellite images of cloud cover, and historical weather information to predict solar output and devise generation and storage dispatch schedules. The same system predicts electricity demand using historical consumption data. The study also notes government initiatives toward a Smart Grid 2.0, alongside measures such as demand response and energy storage systems to enhance resilience. Together, these approaches point to a planning mindset that pairs infrastructure upgrades with digital forecasting and dispatch to manage both rising load and variability.