Stock image.
By displaying energy use per query, ChatNetZero shows that energy use of AI is not necessarily about the underlying model, but rather how the system workflow is designed, and how hallucinations are checked. As such, more complex systems can increase energy consumption without improving search results.
The new update also shows real-world equivalents to the query’s energy usage, such as the power to run household appliances, allowing professionals the ability to more accurately quantify the impact of AI systems.
Launched in 2023, ChatNetZero was created to improve transparency around net-zero targets by allowing policymakers, journalists, investors, and civil society to scrutinise climate pledges made by governments and businesses.
Its outputs are grounded in peer-reviewed datasets and include source citations, reducing the risk of ‘hallucinated’ responses present in other large language models (LLMs). As ChatNetZero is trained on targeted climate datasets and documents, the computational workload required for each query is significantly lower than that of general-purpose AI.
ChatNetZero co-developer, James Zhang of Arboretica, stated: “If AI is to fulfil its promise as a tool for tackling the climate crisis, it must also be accountable for its own environmental footprint. By making energy consumption visible, we’re showing that this level of transparency is possible and meaningful – and calling on other AI companies to follow suit.
“Transparency cannot be optional in the age of AI, especially as these systems increasingly shape public understanding, policy decisions, and investment flows.”
Dr Angel Hsu, co-developer of ChatNetZero and director of Data-Driven EnviroLab, added: “As LLMs become more powerful, they also become more resource-intensive. We redesigned ChatNetZero to move away from monolithic document processing toward a targeted retrieval system that improves efficiency without compromising analytical integrity.
“By dynamically retrieving only relevant materials and disclosing estimated energy use per query, we are strengthening both climate accountability and AI accountability.”
According to the International Energy Agency (IEA), electricity demand from data centres is set to more than double globally by 2030, reaching approximately 945 TWh, driven largely by the rapid adoption of AI.
Related feature: The AI boom risks undermining global energy efficiency efforts
