Artificial intelligence has become a cornerstone of modern business, promising transformative efficiencies across industries. But as AI adoption accelerates, so does its energy consumption. The environmental footprint of AI models, particularly large-scale generative models, remains largely opaque, raising concerns about sustainability in an era of mounting pressure to curb carbon emissions.
A new benchmark for AI sustainability
A new initiative, the AI Energy Score, aims to bring transparency to this issue. Developed by Salesforce in collaboration with Hugging Face, Cohere, and Carnegie Mellon University, the benchmarking tool offers a standardised framework for measuring and comparing AI models based on their energy efficiency. The objective is clear: to create a recognised rating system that incentivises developing and adopting more sustainable AI technologies.
The AI Energy Score, launched at the AI Action Summit, introduces a five-star energy efficiency label to help businesses and developers make informed decisions about the environmental impact of their AI choices. Much like the ENERGY STAR label for household appliances, this system provides an accessible way to compare AI models, rewarding those that deliver performance with lower energy consumption.
Addressing the hidden costs of AI
The initiative arrives at a critical moment. While AI is celebrated for its ability to streamline operations and enhance productivity, its energy demands are rapidly increasing. Research suggests that training a single large language model can consume as much electricity as several hundred households in a year. As companies race to deploy AI-driven solutions, energy consumption risks are becoming a hidden financial and environmental cost. Without clear benchmarks, the industry lacks the tools to assess or mitigate this impact.
The AI Energy Score seeks to change that by providing structured assessments for AI models across ten everyday tasks, including text and image generation. A public leaderboard will rank 166 models, offering developers and enterprises a transparent view of where their technology stands. Salesforce has committed to being the first AI model developer to disclose the energy efficiency of its proprietary models under this framework.
The move is expected to drive competition towards more energy-conscious AI. Historically, the industry focuses on model performance and accuracy with little regard for sustainability. The AI Energy Score challenges this paradigm by positioning energy efficiency as a key differentiator. The French Government and the Paris Peace Forum have already recognised the initiative’s potential to reshape the AI landscape.
Salesforce’s involvement in the project aligns with its broader push towards sustainable AI development. The company has highlighted how its own AI agent platform, Agentforce, is designed with efficiency in mind. By optimising small language models rather than relying solely on large-scale generative AI, Salesforce claims to reduce energy consumption while maintaining performance. This contrasts with more traditional AI approaches that require energy-intensive retraining for each customer, adding to the computational burden.
The future of AI accountability
There is growing recognition that AI regulation and governance must extend beyond ethical considerations to include environmental responsibility. Without accountability, the rapid expansion of AI infrastructure risks deepening global energy challenges at a time when businesses and governments are under pressure to meet net-zero targets. Initiatives like the AI Energy Score mark a step towards aligning AI progress with sustainability imperatives.
However, questions remain. How widely will the industry adopt this new benchmark? Will AI developers willingly disclose their energy consumption data, or will sustainability remain an afterthought in the race for ever more powerful models? As AI integration continues to scale, these questions will define whether innovation can be both intelligent and responsible.
