AI data centers could reduce power draw on demand, study says (www.engadget.com)

A recent trial in the UK has shown that AI data centers can significantly reduce their power consumption on demand, demonstrating the ability to dynamically adjust energy use without disrupting critical workloads. Over five days, more than 200 simulated grid events tested a London data center's responsiveness, revealing that it could cut power draw by up to 40% while maintaining operational stability. Powered by software from Emerald AI and supported by notable partners like NVIDIA and the National Grid, the facility effectively managed peak demand scenarios, such as during soccer match halftimes, where it reduced consumption by 10% for extended periods. This breakthrough holds substantial implications for the AI/ML community, as it positions AI data centers as adaptable participants in energy management, potentially alleviating strain on power grids and lowering operational costs. By establishing a model for a 100MW "power-flexible AI factory" in Virginia, NVIDIA aims to create infrastructure that enhances grid stability and accelerates the deployment of AI workloads without necessitating costly grid upgrades. The collaboration intends to share findings with industry stakeholders and regulators to encourage proactive energy management strategies that could benefit both data center operators and the broader energy landscape.