Concerns about the environmental impact of artificial intelligence (AI) data centres have come to the forefront amid ongoing debates about their necessity and sustainability. A researcher affiliated with the University of British Columbia (UBC), involved in governmental consultations on AI, has expressed skepticism about the need for expanding data centre infrastructure, emphasizing the environmental, economic, and social costs associated with their construction and operation.

The researcher acknowledges that AI holds significant potential benefits, particularly in healthcare. For example, in collaboration with BC Cancer, their team developed a natural language processing (NLP) model designed to analyze pathology reports, identifying high-risk breast cancer patients with an accuracy exceeding 98 percent. This AI-enabled triaging expedites treatment by increasing the proportion of patients receiving presurgical chemotherapy within four weeks from 21 percent to 50 percent. Despite the AI’s supportive role, human pathologists remain responsible for definitive diagnoses.

Importantly, the researcher argues that this type of AI application does not require extensive new data centre capacity. The NLP model consumes minimal computational resources when processing individual patient data — less than one minute of graphics processing unit (GPU) time per patient, equating to less energy than five minutes of video gaming. Though training the model initially demanded significant computational power, comparable to 5,000 hours of gaming, such training occurs infrequently, with maintenance requiring far less energy over time. Even scaling these models to cover multiple diseases would not substantially increase energy demands to the point of justifying new data centres.

The push for expanding data centre infrastructure is largely driven by major multinational technology companies working to develop large language models (LLMs) like OpenAI’s ChatGPT and Anthropic’s Claude. These frontier models require exponentially greater amounts of data and computational power to train, necessitating substantial energy and infrastructure investments. While LLMs have promising applications in healthcare, research suggests that smaller, fine-tuned models can perform healthcare tasks more accurately and with far less resource consumption.

The researcher highlighted a growing tension between private sector ambitions and public interest. While technological advancements in AI have the potential to deliver societal benefits, the rapid expansion of data centres driven by corporate profit motives risks exacerbating environmental degradation, economic inequality, and labour issues, particularly for marginalized communities. The construction and operation of new data centres should be justified by clear public benefits that outweigh such costs.

Looking ahead, the researcher called for rigorous management and regulation of computational infrastructure, likening computational capacity to essential public resources like water and energy. Debates about new data centre proposals are expected to escalate at federal, provincial, and municipal levels. Policymakers and stakeholders are urged to prioritize the public good and national priorities in such discussions to balance technological progress with sustainability and equity concerns.