Healthcare’s integration of artificial intelligence brings significant potential for better patient outcomes, yet it also creates a substantial environmental footprint. AI systems require massive computational power and IT infrastructure, which consumes energy and generates carbon emissions. Dr. Manijeh Berenji, a clinical professor at UC Irvine, identifies three primary drivers of these emissions: the training of AI models, the ongoing process of running models, and the supporting infrastructure like data centers and networks. While model training is energy-intensive, Berenji notes that the longer-term emissions driver is inference, where a deployed tool may process millions of interactions annually.
The Energy Cost of Medical Data
Healthcare AI models tend to produce higher emissions than general-purpose models because they rely on advanced, resource-intensive systems capable of processing detailed medical data. Berenji points out that not all applications carry the same environmental burden. Ambient AI scribes, which transcribe and summarize every patient visit continuously, are among the most energy-intensive uses. Imaging models that store and reprocess high-resolution images also place a heavy load on systems. Conversely, predictive models focused on singular tasks, such as generating sepsis alerts or readmission risk scores, generally use fewer resources.
Organizations must weigh the computational intensity of a tool against the clinical value it provides. A system that increases emissions without improving care works against the medical principle of doing no harm. Berenji suggests treating environmental impact as a key dimension of value alongside safety and cost. For instance, an oncology model that identifies cancer at an earlier stage can reduce the frequency and intensity of treatments, thereby lowering the environmental footprint of care. This efficiency extends to imaging models that eliminate the need for duplicate studies and patient travel.
Decisions made during the initial rollout of electronic health records set the stage for their long-term use. Similarly, the choices made today regarding AI deployment will shape the industry’s environmental impact for years to come. Leaders must ask vendors specific questions about energy use per inference and where workloads are hosted. Framing these environmental concerns as cost-saving measures can also engage executives who prioritize financial metrics over sustainability targets. Collaboration among institutions is essential to prevent redundant efforts and share effective strategies for reducing emissions.
Strategies for Greener AI Deployment
Organizations must design AI workflows for maximum efficiency. Pasting patient information into an AI-powered scribe repeatedly creates redundancy. Teams should avoid running AI on every patient encounter by default. Instead, they should apply the technology only when it provides specific value. Disciplined usage includes scheduling nonurgent workloads like batch analytics during periods of low demand or when clean energy is available.
