As artificial intelligence spreads through daily life and industry, a call for clearer climate accounting is getting louder. Researcher Sasha Luccioni is urging the field to track energy use and user behavior with far more precision, arguing that current blind spots make it hard to judge AI’s environmental cost or reduce it.
“Researcher Sasha Luccioni argues we need better emissions data and a better sense of how people are using AI in the first place.”
Her warning comes as model training grows larger and AI tools reach millions of users across cloud platforms, workplaces, and classrooms. Without reliable data, policymakers, companies, and consumers have little basis to compare systems, set targets, or assess trade-offs.
Why AI’s Carbon Math Remains Murky
AI systems draw power in two major phases: training and use. Training large models can take weeks on specialized hardware. After release, the many queries from users can add up to significant energy use as well. Emissions depend on the energy mix of the data center, the hardware used, and how often people access the system.
Researchers often lack exact figures for any of these factors. Companies frequently treat energy use, server locations, and workload patterns as proprietary. Even when numbers are shared, they may not include enough detail to compare models, adjust for local grids, or account for off-peak scheduling that can cut emissions.
The result is a patchwork of estimates. That makes it hard to tell whether better chips, new code, or cleaner power are actually reducing emissions across the full life of a system.
Growing Pressure to Measure and Report
Public agencies are warning that data center demand is rising quickly. The International Energy Agency estimates that global data centers used several hundred terawatt-hours of electricity in 2022 and could approach double that by the middle of this decade. AI is a growing share of this demand, with everyday use of chatbots, image tools, and code assistants swelling the load.
Some governments are moving to improve transparency. Energy reporting rules for large data centers are advancing in parts of Europe. In the United States, federal guidance and procurement standards are pushing toward higher efficiency and cleaner power for federal workloads. Yet few rules require standardized AI-specific reporting, such as per-model carbon intensity or inference energy per user session.
Industry groups and academics have proposed templates that align with the Greenhouse Gas Protocol. These include reporting the location-based and market-based electricity emissions for training runs, hardware types, utilization rates, and data center regions. They also suggest tracking usage-phase metrics, such as energy per 1,000 queries, to reflect real-world demand.
How Usage Patterns Shape the Footprint
Luccioni’s second point focuses on behavior. Understanding what people do with AI is as important as measuring the servers behind it. A model used for short text responses will have a different energy profile than one serving long video outputs. Peak hours, caching, and batching also change the equation.
Without this usage picture, efficiency efforts risk missing the mark. For example, slimming a model may save energy per query, but if lower costs trigger far more use, total emissions could still rise. Clear data on sessions, prompt length, and response size can guide choices on model scaling, default settings, and pricing that curb waste.
- Track energy for both training and use.
- Report where and when workloads run.
- Measure energy per query and per user session.
- Disclose hardware types and utilization rates.
What Companies and Policymakers Can Do Now
Experts recommend that AI developers publish model cards that include energy and emissions fields, audited by third parties where possible. Cloud providers can offer verified, region-specific carbon data and tools that route jobs to low-carbon hours. Enterprises deploying AI should measure usage intensity and set internal budgets for emissions, not just cost.
Policymakers can set common reporting standards to avoid incompatible disclosures. That would let buyers compare models on both performance and carbon intensity, similar to nutrition labels or appliance ratings. Research funders can require grantees to report training and usage footprints and share methods so others can repeat them.
Balancing Innovation and Impact
Supporters of AI argue that the technology can also drive climate gains, from grid optimization to material discovery. Both views can hold true. Clear, consistent measurement helps separate marketing from measurable savings and points investment to the most effective areas, such as hardware efficiency, model distillation, and clean energy supply.
Luccioni’s message is practical: better data on emissions and use will make choices clearer for everyone. With AI adoption accelerating, the window to build sound accounting into the field is closing fast.
The next steps are straightforward. Set shared reporting rules, publish energy and usage metrics for major models, and align cloud operations with cleaner power. Readers should watch for standardized disclosures from leading labs and cloud providers this year, and for regulators to tie transparency to high-risk or high-compute systems. That is how claims about “efficient AI” move from slogans to facts.
Rashan is a seasoned technology journalist and visionary leader serving as the Editor-in-Chief of DevX.com, a leading online publication focused on software development, programming languages, and emerging technologies. With his deep expertise in the tech industry and her passion for empowering developers, Rashan has transformed DevX.com into a vibrant hub of knowledge and innovation. Reach out to Rashan at [email protected]
