Researchers at the University of California, Riverside have proposed a new framework called Federated Carbon Intelligence (FCI) to reduce the environmental cost of AI. They point out that data centers—where AI workloads run—consume huge amounts of electricity and water, much of which comes from fossil-fuel sources, contributing significantly to air pollution.
The FCI system monitors servers in real time, assessing their temperature, age, and wear, and combines this with data on how carbon-intensive electricity is at different times and locations. With this information, it intelligently routes AI tasks to servers that both minimize emissions and reduce hardware stress, optimizing for both sustainability and longevity.
In simulations, FCI was shown to potentially cut carbon dioxide emissions by up to 45 percent over five years, while also extending the operational life of server hardware by about 1.6 years. By avoiding overworking stressed machines, the system can lower the need for intensive cooling, reduce failure rates, and delay costly replacements.
The researchers argue that sustainability in AI can’t rely only on cleaner energy — it also requires smarter use of existing infrastructure. They envision FCI being deployed in real data centers without needing new hardware; instead, it would work through better coordination and optimization across current systems, potentially laying the groundwork for a more climate-aligned AI infrastructure.
