Data Center Power Use to Double by 2030, Straining Grids

Global data center electricity consumption is projected to reach 565 TWh in 2026, marking a 26 per cent year on year increase, according to a recent Gartner report. This demand is straining power infrastructure worldwide.

Credit: UNSPLASH

By 2030, this figure is forecast to exceed 1,200 TWh. This amount surpasses Japan's entire annual electricity consumption. Concurrently, at least 75 data center projects valued at USD 130 billion were either postponed or cancelled across the United States. These delays stemmed from a lack of sufficient power, rather than issues with funding or semiconductor availability.

And Microsoft Chief Executive Officer Satya Nadella recently voiced concerns on a podcast, stating that the principal challenge is no longer excess compute capacity. The difficulty now lies in deploying power swiftly enough to data facility locations. He warned that without a solution, chips would remain in warehouses, unable to function.

Figures show a substantial rise in AI optimized server electricity consumption. It is projected to jump from 95 TWh in 2025 to 175 TWh in 2026, representing an 84 per cent surge. By 2027, the power usage of AI servers is expected to officially exceed that of traditional servers. A single graphics processing unit server consumes three to five times the power of a standard server.

But Big Tech companies are increasing infrastructure spending significantly. The combined capital expenditure by Microsoft, Amazon, Google, and Meta is anticipated to reach approximately USD 650 billion in 2026. A substantial portion of this funding is allocated to AI infrastructure. Microsoft alone spent USD 120 billion in 2025. Amazon forecasts USD 118 billion for the same period, while Google's capital expenditure is estimated at more than USD 850 billion.

The International Energy Agency projects global data center electricity demand will more than double from 2024 levels. It expects this demand to reach approximately 945 TWh by 2030. The United States and China together are forecast to account for 80 per cent of this demand. In China, the rapid expansion of AI computing capacity is placing new pressure on an already strained power grid infrastructure.

So, a report by Capgemini Research Institute found that 80 per cent of surveyed power executives predict data center peak loads will become more extreme and less predictable. This report surveyed 612 power industry executives and 175 data center executives. Furthermore, 79 per cent of respondents rated this situation as a severe challenge.

Tech giants are addressing these issues directly. Google has announced plans to restart the Duane Arnold Energy Center nuclear plant in Iowa. This facility had been closed since 2020. The target for recommissioning is 2029. In China, the government is pursuing a coordinated approach to balance AI infrastructure growth with grid capacity, investigating nuclear, renewable, and more efficient cooling solutions.

And the underlying message is clear: AI demands have outpaced existing power grids. Building new power capacity is now as crucial as developing advanced semiconductor chips.

• Global data center electricity consumption is forecast to reach 565 TWh in 2026, increasing to over 1,200 TWh by 2030.

• The escalating power demand for AI infrastructure has led to the postponement or cancellation of numerous data center projects in the United States.

• AI optimized servers are projected to use 175 TWh by 2026, surpassing traditional server power consumption by 2027.

• Major tech companies are committing significant capital expenditure to AI infrastructure, with figures exceeding hundreds of billions of USD.

• Governments and tech organisations are exploring nuclear, renewable, and efficient cooling solutions to address grid capacity issues.

Source: Pandaily