- AI data center power consumption will soon exceed that of conventional data centers
- Global data center power consumption has increased by 26% since 2025
- In the US, AI data centers account for 36% of all data center power consumption
Data centers built to add capacity for AI will soon consume more power than conventional data center hardware. With consumption at 175 terawatt-hours (TWh) in 2026, predictions take that figure to 258 TWh in 2027 – where AI-optimized data centers will overtake conventional data centers for consumption.
Compared to 2025, AI data center power consumption has increased by about 84%. These are the results of a Gartner forecast that also predicts AI-optimized servers to account for 31% of data center power consumption in 2026.
This level of consumption, combined with an excess of demand for electricity relative to production, will be the main constraint on future AI expansion, Gartner predicts. But data center consumption as a whole is expected to be 565 TWh in 2026, up 26% year over year.
Consumption is increasing, but capacity is growing slowly
When it comes to global consumption, the US accounts for 36% or about 204 TWh of the full 565 TWh of global demand. Within that portion of US demand, AI data centers account for a third, with consumption expected in 2026 to be around 68 TWh.
“Increasing demand for compute-intensive AI workloads is driving unprecedented growth in data center power, while AI capacity is now limited by power availability, making data center power security the new battleground for scaling and protecting margins in the global AI race,” said Gartner direct analyst Linglan Wang.
By 2030, Garter predicts that supply will not be able to meet demand when consumption passes the 1,200 TWh mark.
To address this limitation, Wang suggested that business leaders and infrastructure providers should focus on upgrading the efficiency of power grids and the hardware that draws the most power, such as cooling systems.
A recent Google Cloud report further suggested that to address the rising cost of power consumption, companies should move from running AI models on a centralized cloud to edge deployments, where the efficiency of these systems is increased with the added benefit of avoiding a global disruption of services if the centralized cloud system fails.
But these levels of spending increases are unlikely to cool the growing anti-data center sentiment in the US, which – combined with hardware and power generation shortages – has seen nearly half of all data centers by 2026 delayed or cancelled.
Via Tom’s hardware
Follow TechRadar on Google News and add us as a preferred source to get our expert news, reviews and opinions in your feeds.
