InvestorWaves
Latest stories

Data Centers Are Becoming Power Hungry

Get more data-driven visuals in your inbox.

Chart for Data Centers Are Becoming Power Hungry
Data Centers Are Becoming Power Hungry

Data centers are turning into one of the biggest new power-demand stories in the market.

The IEA estimates that global data-center electricity consumption was about 415 TWh in 2024. In its Base Case forecast, that rises to about 945 TWh by 2030, more than doubling in six years. A lot of that increase is tied to AI workloads, which require more computing power, denser server racks, more cooling, and more reliable grid connections than the last generation of cloud infrastructure.

For investors, the number is useful because it connects the AI trade to the physical economy. The demand is not only showing up in chip orders or cloud capex. It is also showing up in electricity load, utility planning, transmission needs, backup power, cooling systems, transformers, and data-center construction. That widens the AI infrastructure theme beyond the obvious software and semiconductor names.

The forecast also helps explain why power has become a bottleneck in some markets. Building a data center is not just a real estate project. The site needs enough electricity, the grid needs enough spare capacity, and local utilities need enough time to connect the load. As AI clusters get larger, those constraints can decide where new capacity is built and which suppliers benefit.

The key caveat: 2030 is a forecast, not a reported number. The final path depends on AI adoption, chip efficiency, model training demand, inference demand, cooling improvements, and how quickly utilities can expand capacity. Still, the direction is clear in the IEA data: data centers are becoming a much larger part of global electricity demand.

Source: IEA Energy and AI
Back to latest stories

Market insights, made visual.

Get data-driven insights into stocks and investing themes in a free newsletter that takes 5 minutes to read.

Join free. Unsubscribe anytime.