Is AI Killing Polar Bears?
Learn how AI’s energy use and data-center emissions may contribute to climate change, and what that means for polar bears. The answer separates AI’s indirect climate impact from the direct causes of polar bear decline.
Overview
AI is not directly killing polar bears, and no study has measured a number of bear deaths caused by AI. There is, however, a plausible indirect pathway: AI uses electricity in data centres; where that electricity causes additional greenhouse-gas emissions, it adds to human-driven warming; warming reduces Arctic sea ice; and bears depend on sea ice to hunt seals. The first and last links in that chain are well established, but the fraction of warming or polar-bear decline attributable specifically to AI has not been quantified.
The clearest answer to “does AI kill polar bears?” is therefore: not directly, but AI-related energy demand can contribute at the margin to the climate pressure that threatens their habitat. AI can also support wildlife monitoring and energy efficiency, so its overall climate effect depends on how it is powered, used, and governed.
How AI could affect polar bears
AI models are trained and run on servers, mostly in data centres. Those facilities use electricity for computing, storage, networking and cooling. If added demand is met by fossil-fuel generation, it produces additional carbon dioxide emissions. The causal pathway to bears is indirect and cumulative—not a mechanism in which an AI system encounters or harms an animal.
The diagram shows a conditional pathway. Electricity emissions depend on the power supply and whether the demand is genuinely additional. Warming and sea-ice loss have multiple human and natural influences; available evidence does not isolate AI’s contribution to Arctic change or to individual bear deaths. The AI and climate change - Energy and AI - IEA estimates emissions for all data-centre workloads, with AI only one part of that total.
How large is the electricity demand?
The IEA estimated that data centres used about 415 terawatt-hours (TWh) of electricity in 2024, approximately 1.5% of global electricity consumption. In its base-case projection, total data-centre electricity demand reaches about 945 TWh in 2030. AI is a major driver of projected growth, but these figures cover all data-centre workloads—not AI alone. The IEA also stresses uncertainty about AI adoption, efficiency improvements and energy-system constraints. Executive summary - Energy and AI - IEA
The IEA’s 2025 assessment puts electricity-related data-centre emissions at around 180 million tonnes of CO₂ today, and projects these to rise in its base case. Again, this is for the whole data-centre sector, not AI on its own. The electricity mix matters: the IEA expects renewables to supply nearly half of data-centre demand growth through 2030, while natural gas and coal together meet more than 40% of the additional demand in its base case. Renewable generation and efficiency can reduce the emissions associated with a given workload, but they do not establish that the workload has no climate impact. AI and climate change - Energy and AI - IEA; Energy supply for AI - Energy and AI - IEA
These figures describe infrastructure-scale pressure, not a polar-bear mortality estimate. AI-specific accounting is difficult because facilities combine workloads and because emissions depend on location, time of use, grid generation, equipment, cooling and accounting methods. No defensible calculation currently converts an AI query, data-centre expansion or quantity of AI-related CO₂ into a number of polar bears killed.
Why sea ice matters to polar bears
Polar bears (Ursus maritimus) use sea ice as a platform for hunting seals, their principal prey. When ice cover shrinks or the ice-free season lengthens, bears can lose hunting opportunities, spend longer periods fasting on land, and face greater difficulty maintaining the body condition needed for survival and reproduction. The U.S. Geological Survey identifies reduced access to prey as the main expected pathway from sea-ice loss to polar-bear effects, while also noting possible changes in prey distribution and energy expenditure. Health and Energetics of Polar Bears | U.S. Geological Survey
A 2025 Science study analysed 42 years of Western Hudson Bay monitoring data using a model of individual bears’ energy intake and expenditure. The researchers reported that climate-driven sea-ice loss and reduced hunting opportunity explained an approximately 50% decline in that subpopulation from 1979 to 2021. The model linked shorter hunting periods to declining body mass, fewer cubs and lower survival. This is strong, population-specific evidence for the sea-ice mechanism; it does not show that AI caused the decline, nor can one regional result be treated as a uniform trend for every polar-bear population. New Study Quantifies Link Between Sea Ice and Polar Bears | Polar Bears International
Regional variation matters. At its 2024 working meeting, the IUCN/SSC Polar Bear Specialist Group recognized 20 subpopulations: 10 were data deficient, three likely decreasing, five likely stable and two likely increasing. The group still identified sea-ice loss as the primary threat to the species’ persistence. Those classifications show why a single global statement that every polar-bear population is already declining would be inaccurate. 20th Working Meeting of the IUCN/SSC Polar Bear Specialist Group - Polar Bear Specialist Group
Recent Arctic sea-ice measurements
NOAA’s 2024 Arctic Report Card records September 2024 average sea-ice extent at 4.38 million square kilometres, compared with a 1991–2020 September average of 5.58 million square kilometres. September 2024 ranked sixth-lowest in the 45-year satellite record. Sea-ice extent is the area with ice concentration of at least 15%; it is not a direct measure of bear numbers or of ice quality in each subpopulation’s habitat. Report Card 2024 - NOAA Arctic; NOAA Arctic Report Card: Sea Ice
The U.S. Fish and Wildlife Service continues to classify polar bears as threatened under the U.S. Endangered Species Act. It says polar bears rely heavily on sea ice and that rising greenhouse-gas concentrations contribute to Arctic warming and continued sea-ice habitat loss. Its review gives an approximate global population estimate of 26,000 (95% confidence interval: 22,000–31,000), but that estimate is based on cited 2015 research and should not be read as a precise, current census. Polar Bear 5-Year Status Review Frequently Asked Questions | U.S. Fish & Wildlife Service
What the evidence can—and cannot—establish
| Question | Evidence-based answer | Important limit |
|---|---|---|
| Does AI directly kill bears? | There is no established direct mechanism or documented AI-attributed death count in the evidence reviewed. | Absence of a quantified case is not a claim that AI has no environmental costs. |
| Can AI add to climate pressure? | Yes, where AI adds electricity demand that results in additional greenhouse-gas emissions. | AI is only a portion of data-centre activity; electricity and emissions attribution varies. |
| Is climate-driven sea-ice loss harmful to bears? | Yes. It restricts access to seal prey and can reduce energy reserves, reproduction and survival. | Effects differ among subpopulations; some population estimates are old or data deficient. |
| Has AI been shown to cause a specific population decline? | No. | The 2025 Western Hudson Bay study attributes the population decline it examined to climate-driven sea-ice loss, not to AI specifically. |
The best-supported statement is that AI-related emissions are a possible incremental contributor to a much larger climate problem, and climate-driven sea-ice loss is a well-established threat to polar bears. It would overstate the evidence to say AI is “killing polar bears” as if researchers had traced the deaths to AI, or to attribute a particular share of the observed declines to AI. It would also be wrong to infer from that uncertainty that the extra demand is harmless: the scale and carbon intensity of new electricity supply affect the direction and size of its contribution.
Other factors can affect bears locally. The Fish and Wildlife Service discusses harvest, contaminants, disease and other stressors, while the Polar Bear Specialist Group notes concerns about human–bear conflict as food-stressed bears spend more time near communities and camps. These issues can be important for management and particular populations, but they do not negate the central conservation concern about sea-ice habitat. Polar Bear 5-Year Status Review Frequently Asked Questions | U.S. Fish & Wildlife Service; 20th Working Meeting of the IUCN/SSC Polar Bear Specialist Group - Polar Bear Specialist Group
AI can also help monitor polar bears
The climate footprint of AI is not its only environmental role. Computer-vision systems can help researchers screen large collections of aerial, satellite or camera images for wildlife. A 2021 peer-reviewed study trained a convolutional neural network using web images and aerial-survey background images to detect polar bears on sea ice. In its test data, it correctly classified 20 of 21 bear images (95%) and 1,172 of 1,179 background images (99.4%). Those results demonstrate a possible way to speed image review; they do not establish the accuracy of a full population census or guarantee equal performance in different conditions. Using Web images to train a deep neural network to detect sparsely distributed wildlife in large volumes of remotely sensed imagery: A case study of polar bears on sea ice
Automated detection can reduce the time needed to locate animals in imagery and may complement conventional field surveys. It cannot by itself determine population trends, replace local knowledge, or remove the need for validation: image resolution, snow and ice conditions, viewing angle and the model’s training data all affect detection. The Polar Bear Specialist Group emphasizes the continuing need for diverse scientific methods, location data and collaboration with Indigenous communities for conservation and management. 20th Working Meeting of the IUCN/SSC Polar Bear Specialist Group - Polar Bear Specialist Group
AI may also reduce emissions in some applications. The IEA identifies potential uses in renewable-power forecasting, industrial energy efficiency, transport and building controls. It estimates that widespread adoption of existing AI applications could produce emissions reductions larger than data-centre emissions in its scenario, but warns that the savings depend on deployment and enabling conditions and could be reduced by rebound effects. These are modeled potential benefits, not a verified offset for the emissions of any particular AI service or data centre. AI and climate change - Energy and AI - IEA
What would reduce AI’s indirect climate contribution?
The relevant measures address the infrastructure and energy system rather than an individual bear–AI interaction. Lower-carbon electricity, more efficient chips and models, efficient cooling, careful facility siting and transparent energy and emissions reporting can reduce the climate impact per unit of computing. The IEA’s analysis underscores that grid mix, system efficiency and the pace of demand growth all affect emissions; renewables are expanding, but fossil generation is also expected to supply part of data-centre growth. Efficiency matters, yet cheaper or more capable computing can stimulate additional use, so efficiency alone does not guarantee lower total energy demand.
For polar bears, the direct conservation priority remains limiting greenhouse-gas-driven warming and preserving the sea-ice habitat needed for hunting, alongside region-specific population monitoring and management. AI has a role both as a source of additional electricity demand and as a potential research and efficiency tool. Its net effect is not predetermined, and current evidence cannot assign AI a share of bear deaths.
Conclusion
AI is not known to kill polar bears directly. It can contribute indirectly if its growing electricity demand causes additional greenhouse-gas emissions, which add to warming and sea-ice loss—the primary habitat-related threat to bears. The climate-to-sea-ice-to-bear mechanism is supported by ecological research and observations; the AI-specific contribution to that chain remains unmeasured. Meanwhile, AI-assisted image analysis can aid monitoring, and carefully implemented AI applications could reduce emissions elsewhere. The accurate conclusion is neither that AI is harmless nor that it is a proven direct killer: its marginal effect depends on the energy and climate consequences of how computing is produced and used.