New Distributed Data Center Model Proposed by Tech Companies; Utah Mega-Project Approved Amid Scrutiny

The escalating demand for artificial intelligence computing power is pushing electricity supply limits, with the U.S. Department of Energy forecasting a near-tripling of data center electricity usage by 2028. This trend is exemplified in Utah by the proposed Stratos Project in Box Elder County, which, backed by Kevin O’Leary and approved by the Military Installation Development Authority, could consume up to 9 gigawatts—more than double Utah's current power use. This immense power demand has prompted residents and state officials to call for increased scrutiny before the project advances. In response to the high costs, complex infrastructure requirements, and delays associated with traditional large data centers, a trio of companies is exploring a distributed data center model. Span, a San Francisco-based company known for its smart electrical panels, is spearheading this initiative. They are collaborating with Nvidia, providing the computing chips, and homebuilder PulteGroup to pilot small outdoor AI compute nodes, known as XFRA units, in approximately 100 newly constructed homes. These XFRA units are designed to handle AI inference workloads, leveraging existing residential electrical capacity to offer a more rapid and cost-effective solution than conventional centralized data centers. Homeowners would receive the necessary equipment, including a Span electrical panel and battery backup, at no upfront cost, with the potential benefit of reduced electricity and internet bills. However, this novel approach introduces challenges concerning noise, heat, maintenance, and liability for homeowners, and necessitates a regulatory framework to determine whether these backyard AI units should be classified as household appliances or commercial infrastructure.