Skeleton Technologies develops new GrapheneGPU battery for AI data centers

AI data centres are consuming more electricity, and grid infrastructure is struggling to keep pace. The International Energy Agency expects global AI data centre electricity demand to reach 945 TWh annually by 2030, with over 150 TWh coming from Europe.

One technical challenge is the erratic power draw of GPUs, which can spike from zero to full load within seconds. Many data centres deal with this by using dummy loads during idle periods. This helps stabilise the power draw but wastes energy. According to Talinn-based Skeleton Technologies, up to 45 percent of energy can be lost as heat without producing any computation.

The well-funded startup, backed by the likes of Siemens, has introduced a new battery system called GrapheneGPU, designed to reduce these losses and support data centers. Built with the company’s patented ‘Curved Graphene’ material, the system stores energy during idle times and releases it during peak demand. This reduces the need for dummy loads and smooths out fluctuations in power consumption. The design is also made by a German factory, the company says.

The company claims the system can reduce AI system energy use by up to 45 percent and lower peak power demand by 44 percent. It also reports a 40 percent increase in computing throughput, measured in FLOPS, under test conditions. These results come from validation under real GPU load profiles. The battery is certified for peak shaving in data centres and is compatible with existing infrastructure.

Skeleton says the first units will ship from its German production site in June 2025. A second manufacturing line in the US is planned for early 2026.

The system combines its supercapacitors with software that uses AI to manage charging and discharging.

From pv magazine Germany.