Nebius: How to Make AI Infrastructure Less Energy Intensive With investments in AI infrastructure going stratospheric, attentions are increasingly turning to the energy systems needed to support the next generation of data centres. For operators building facilities capable of handling ever-larger AI workloads, improving computational performance is only part of the equation. Reducing electricity demand, lowering water consumption and integrating sites with local energy networks are becoming equally important priorities. Nebius believes the answer lies in designing its AI cloud as a fully integrated system, combining servers, racks, cooling technologies and cloud software to maximise efficiency. The company's latest Sustainability Report outlines how this engineering-led approach is helping reduce the energy and resource intensity of its expanding global infrastructure while supporting rapid growth. Reducing electricity demand through integrated design With AI driving unprecedented growth in data centre electricity consumption, improving operational efficiency has become a central challenge for the sector. Nebius reports that its global data centre portfolio achieved an average Power Usage Effectiveness (PUE) of 1.25 in 2025, comfortably below the reported global industry average of 1.54. According to the company, its combination of purpose-built hardware and advanced cooling technologies prevented around 102GWh of electricity consumption over the course of the year. Nebius says that figure is comparable to the annual electricity consumption of approximately 9,800 households in the US. "How AI is built matters just as much as what it enables," says Daria Mukhortova, Head of Sustainability at Nebius. "Our job is to clearly report the impact of increasing demand for AI, and to use the data to improve how we build. Since our inception, we have measured and reported our impact, because transparency is how we stay accountable as we scale." Rather than treating servers, racks and cooling systems as separate components, Nebius develops them together as
Nebius: How to Make AI Infrastructure Less Energy Intensive
Read the original article
energydigital.com →