AI Models Hiding Their Energy Footprint? Here’s What You Can Do
What is the AI Energy Score? ⚡
The AI Energy Score provides:
- Standardized Energy Ratings: A uniform benchmark measuring AI models’ energy consumption for specific tasks, ensuring clear and reliable evaluation of model efficiency.
- Public Leaderboard: Our leaderboard currently ranks 166 AI models across 10 key tasks such as text and image generation, summarization, and more. Featured models include popular open-source ones like LLaMa, Phi, Gemma, Mistral, and SmolLM.
- Submission Portal: Our approach porta enables both open-source and proprietary models to be submitted for evaluation, ensuring reproducible results and fair comparisons.
- Recognizable Energy Use Labels: Similar to household appliance ratings, our simple 1- to 5-star system guides users toward sustainable AI choices.
The launch had a very positive reaction with widespread interest, earning recognition from prominent outlets like BBC, NPR, Newsweek, Nature, and The Economist. Yet, to unlock the project's full potential, we need greater inclusion of closed-source (proprietary) AI models, currently underrepresented on the leaderboard.
Despite our secure Docker-based submission process, closed AI model providers have yet to submit their in-production models for testing, or share this data in any other way. This reluctance means that the ranges established by the project don't span the full range of models used in production, making objective comparisons impossible. According to OpenRouter, 15 out of the top 20 most-used models this month are closed-source, highlighting a significant gap in the understanding of real-world AI energy impacts. Transparency, even in competitive industries like food and pharmaceuticals, has set a valuable precedent demonstrating that partial openness is both achievable and beneficial.
Given the reluctance of closed AI model providers to participate in the AI Energy Score,** we believe the most effective path forward is to apply pressure through their customers**, especially enterprises, since they hold significant influence and can play a pivotal role in incentivizing providers to move toward greater transparency.
The power of enterprise customers
Enterprises are uniquely positioned to drive change due to their purchasing power and market influence. By explicitly prioritizing transparency and sustainability in procurement decisions, enterprises can compel AI providers to disclose critical data, shifting market norms.
Here’s how your organization can lead:
- Set clear sustainability priorities during procurement processes: Make it clear to AI model providers that sustainability and transparency are key business priorities
- Advocate for industry-wide sustainability transparency: Encourage transparency regulations and voluntary standards.
- Integrate environmental transparency into Tenders/Requests for Proposals (RFPs) and contracts: Require AI solutions to benchmark their models using the AI Energy Score approach and to achieve specific benchmarks by a certain timeline.
Example Tender/RFP Question
As part of our commitment to sustainability and transparency, we ask vendors to disclose the energy efficiency of AI models and systems provided under this RFP. Please provide the AI Energy Score (or an equivalent industry-standard energy efficiency rating) for each AI model or system proposed. If an AI Energy Score is not available, include:
- Estimated energy consumption (in kWh) for typical workloads.
- Carbon footprint per inference/training session.
- Energy optimization techniques used in model training and inference.
- Certifications, benchmarks, or third-party validation related to energy efficiency. Vendors demonstrating a commitment to minimizing AI-related energy consumption and carbon footprint will be prioritized in our evaluation process.
Example Procurement Contract Requirements
Clause: AI Energy Efficiency & Disclosure
- Energy Transparency: The Supplier agrees to provide an AI Energy Score (or an equivalent standard) for all AI models delivered under this contract. If unavailable, the Supplier must disclose energy consumption metrics, including power usage and carbon emissions per AI processing unit (e.g. query), and the energy-saving measures put in place.
- Sustainability Commitment: The Supplier shall make commercially reasonable efforts to improve the energy efficiency of AI models and implement optimizations to reduce energy consumption without compromising performance.
- Ongoing Reporting: The Supplier shall, upon request, provide updated AI Energy Scores or equivalent benchmarks for AI models used in ongoing services. If material changes affect energy efficiency, the Supplier must notify the Buyer and offer an updated version where feasible.
- Compliance & Penalties: Failure to provide required energy efficiency disclosures or a material deviation from the disclosed AI Energy Score (or equivalent) may be considered a breach of contract, subject to corrective measures, renegotiation of terms, or penalties as determined by the Buyer. These requirements align with our organization's sustainability goals and ensure vendors contribute to reducing the environmental impact of AI deployments.
Beyond procurement—additional enterprise actions:
- Establish internal sustainability benchmarks prioritizing energy-efficient AI.
- Support policy initiatives promoting mandatory disclosures and accountability.
- Educate internal teams on the value of AI energy efficiency and sustainability.
Take Action Now 🚀
Your organization's purchasing decisions are powerful levers for change. Insist on energy transparency from AI providers to drive sustainability. Together, let's ensure AI development aligns with our planetary boundaries.
Explore the AI Energy Score Leaderboard today and help create a sustainable AI future!
