New system to cut AI carbon and protect servers ^{CEFR B2}

As AI demand grows, so do the energy and water needs of the data centers that run it, and those centers often rely on electricity from fossil-fuel power plants. Researchers at the Marlan and Rosemary Bourns College of Engineering, University of California, Riverside, propose a new framework called Federated Carbon Intelligence (FCI) to address both operational emissions and hardware health.

FCI links environmental and grid carbon-intensity data with real-time measures of individual server condition. By routing AI workloads with awareness of server temperature, age and physical wear, the system aims to lower carbon emissions while reducing hardware stress. In simulations the researchers report up to a 45 percent reduction in CO2 over five years and an increase in average server fleet life by about 1.6 years.

The method also considers embodied carbon from manufacturing new machines, so keeping servers in service longer helps lower total emissions. Published in MRS Energy and Sustainability, the study’s next step is to partner with cloud providers and test FCI in operational data centers.