Sasha Luccioni 是一位 AI 永续发展研究者，她在 Hugging Face 工作了 4 年；她认为，让 AI 变得可持续，取决于更完善的排放数据，以及更清楚了解 AI 实际是如何被使用的。她正与前 Salesforce 永续主管 Boris Gamazaychikov 一起成立 Sustainable AI Group，时值大型科技公司正在扩建以化石燃料供电的资料中心，而 Trump 政府也在放松环境规则；尽管如此，客户、员工和董事会成员对透明度的要求却日益增加。

在她看来，公司现在需要量化 AI 的足迹，而不是争论是否要使用 AI。她说，企业正被推动去辨识模型运行在哪里、它们依赖哪些资料中心和电网，以及涉及哪些供应链与运输排放。她也希望有产品层级的揭露，例如在 ChatGPT 或 Claude 中加入仪表或资讯框，显示能源使用量，并且最好还能显示温室气体排放以及每次查询背后的能源来源；同时她指出，token 计数已经能帮助公司为简单任务选择更简单的模型，并为更深入的工作保留更大的模型。

Luccioni 说，区域性的政策压力正在上升：欧洲的 EU AI Act 已经纳入永续条款，International Energy Agency 也正在报告 AI 与能源使用，一些国家则因资料中心建设者缺乏 5 年产能规划所需的基本数据而加以抵制。她认为，如果一家大型供应商承诺使用可再生能源供电的资料中心，可能会取得市场优势，但她也警告不要夸大影响，因为单次查询可能很小，而总体使用量会随著使用者数量而扩大。她的主要保留意见是，在任何人能对 AI 真正的环境成本做出有根据、细致的决策之前，研究人员需要可验证的能源与用水数据。

Sasha Luccioni, an AI sustainability researcher who spent 4 years at Hugging Face, argues that making AI sustainable depends on far better emissions data and a clearer picture of how AI is actually used. She is launching Sustainable AI Group with former Salesforce sustainability chief Boris Gamazaychikov at a time when major tech firms are expanding fossil-fueled data centers and the Trump administration is easing environmental rules, even as customers, employees, and board members are increasingly asking for transparency.

In her view, companies now need to quantify AI’s footprint rather than debate whether to use AI at all. She says firms are being pushed to identify where models run, which data centers and grids they rely on, and what supply-chain and transportation emissions are involved. She also wants product-level disclosure, such as a meter or info box in ChatGPT or Claude showing energy use and ideally greenhouse gas emissions and the energy source behind each query, while noting that token counts already help companies choose simpler models for simple tasks and reserve larger models for deeper work.

Luccioni says regional policy pressure is rising: Europe’s EU AI Act already includes sustainability clauses, the International Energy Agency is reporting on AI and energy use, and some countries are pushing back on data-center builders because they lack basic numbers needed for 5-year capacity planning. She believes a big provider that commits to renewable-powered data centers could gain a market edge, but she also warns against overstating impact because individual queries may be small while total use scales with the number of users. Her main caveat is that researchers need verifiable energy and water data before anyone can make informed, nuanced decisions about AI’s real environmental cost.