Greening AI Inference with Accuracy and Latency-aware User Incentives

A newly proposed framework aims to reduce the carbon footprint of artificial intelligence services by offering users financial incentives to accept lower quality or slower responses during periods of high carbon intensity, according to a paper submitted to arXiv in May 2026 [1]. The framework, detailed in a preprint titled "Greening AI Inference with Accuracy and Latency-aware User Incentives," addresses the growing environmental cost of AI, where the inference phase — the process of a trained model generating answers to user queries — has been identified as a major contributor to carbon emissions [1][2]. The widespread adoption of generative AI tools such as ChatGPT, DALL-E, and Midjourney since the 2020s has been supported by large-scale data centers with high energy consumption that is estimated to be growing steadily [3]. The authors propose a practical two-tier service subscription model [1][2]. Under this system, users could opt into a discounted service tier. In exchange for a lower price, the AI provider would gain the flexibility to serve a percentage of inference requests at reduced quality and increased latency specifically when the carbon intensity of the electrical grid is high [2]. The design is built on a formal framework that accounts for a user's individual valuation of inference quality and latency, as well as their personal environmental consciousness [1][2]. This approach taps into established principles of consumer behavior, a field that studies how individual preferences, emotions, and external cues shape purchasing decisions [5]. By making the environmental cost a direct factor in a user's service choice, the framework creates a market mechanism for sustainability. The study notes that the specific tradeoff between carbon emissions and the two quality-of-experience parameters — quality and latency — can vary depending on the size and complexity of the AI models and how resources are allocated to serve requests [1][2]. The paper does not provide empirical data from a live deployment but offers a theoretical structure for AI providers to design such programs. The research comes as the environmental impacts of AI, including electronic waste and fresh water consumption for cooling, face increasing scrutiny alongside the technology's rapid expansion into sectors from software development to healthcare and finance [3].