AI systems require huge amounts of electricity, which dramatically increases the requirements for data center backup power systems in case of failures, writes EE Times. According to media reports, global demand from critical IT infrastructures will grow from 49 GW in 2023 to 96 GW by 2026, with AI accounting for about 40 GW. At the same time, the capacity of the AI data center is expected to be 100 MW. And the total consumption of data centers could grow to more than 1000 TWh in 2026.
AI data centers require 99.99999% availability, so they need full-scale power backup. For this, diesel or gas generators with a capacity of 1–2 MW are usually used. To protect a large data center, dozens of such machines are needed, ready for launch. The speed of switching to a backup power source is critical – it usually takes up to several minutes to start the generator, so all this time the power must be supplied from batteries. In addition, you need a switchgear, as well as a monitoring and automation system that will make the transition from one source to another imperceptible for the IT load.
Image Source: Saif71.com/Unsplash
However, the growth in data center capacity dramatically complicates and increases the cost of creating backup power systems. A megawatt-class generator weighs about 5 tons (without fuel), takes up quite a bit of space (add here a fuel tank with a volume of 1 m3) and costs about $1–$2 million without taking into account delivery and installation. At the same time, the battery system is another separate complex with its own installation problems. So it is possible that the recently launched xAI AI cluster, which is currently forced to be powered by generators, will leave these generators on the site even after connecting to the power grid.