Skip to main content

What is the environmental impact of AI?

The AI Glossary

In our classes, we often pause to explain the ideas behind AI and the language used to describe it. These posts grew out of those conversations. They’re for anyone who would like to understand AI better, with no technical background assumed.

AI’s environmental impact comes from the physical systems needed to build and use it: manufacturing equipment, powering computation, cooling data centers and eventually replacing or disposing of hardware. The “cloud” is not weightless. It is a collection of buildings, chips, cables and energy systems.

Training and inference are different parts of that footprint. Training develops a model by adjusting its learned settings. Inference uses the trained model to answer a request or make a prediction. Training a large model can require substantial resources, while a widely used service may accumulate a larger footprint through repeated inference. Which dominates depends on the model, its use and the period being assessed; there is no universal crossover point.

The calculation also depends on what is counted. Operational emissions arise from running the system. Embodied emissions arise from manufacturing and other stages before and around operation. Looking only at the electricity for one model call leaves out some of the lifecycle. Conversely, assigning all data-center growth to AI would also be misleading, because data centers support many other activities.

Location matters. The emissions associated with electricity vary with the local generation mix and time of use. Cooling arrangements differ, as does local water availability. Water withdrawal—the amount taken from a source—is not the same as consumption, the portion not promptly returned to that source. A single “water per prompt” figure hides these assumptions and should not be presented as a universal property of AI.

Efficiency can reduce the resources needed for a task. Smaller models, better hardware and less unnecessary generation may help. But cheaper computation can encourage more use, partly or fully offsetting those savings. This is a rebound effect. Jevons paradox describes the stronger case in which increased use more than offsets the efficiency improvement; it is a possibility to examine, not an inevitable law for every technology change.

Environmental health also extends beyond carbon. Electricity generation and backup power can affect air quality, and water demand can matter locally. Assessing those effects requires information about the particular infrastructure and affected communities, rather than assuming that every data center has the same impact.

A health organization can begin with a practical comparison: what resources does the complete workflow use, and what benefit does it produce? A smaller model or fixed software rule may be sufficient for routine extraction. A more capable model may be justified where it produces a meaningful improvement. AI can also support environmental research and operational efficiency, but those benefits need measurement. The sensible aim is neither to treat every prompt as an environmental emergency nor to ignore the infrastructure. It is to make resource use visible and proportionate to the work being done.

Sources: IEA — Key Questions on Energy and AI (2026); Wu and colleagues — Sustainable AI: Environmental Implications, Challenges and Opportunities (2022); MIT CS3 — Air Quality and Health Impacts of Data Center Electricity Demand in the United States

To learn more, take a look at the AI in Health Care Certificate of Specialization from the Advanced Learning Academy at Harvard T.H. Chan School of Public Health. You can take an individual course or work toward the full certificate.


AI is a fast-moving field. This explanation reflects our class discussions and related material at the time of writing. Terminology, capabilities and tools may change.

For suggested additions or corrections, please email me at tpanch@hsph.harvard.edu.

Dr. Trishan Panch
Program Director, AI in Health Care
Harvard T.H. Chan School of Public Health — Advanced Learning Academy

Lead with AI in Health Care

The AI Pragmatist is a monthly newsletter from ALA faculty, with new explainers & essays from the Advanced Learning Academy