Strategic Intelligence: Deep Dive into the Environmental Impact of Small Language Models

Strategic Intelligence: Deep Dive into the Environmental Impact of Small Language Models

Summary

The artificial intelligence (AI) industry has been dominated by large language models (LLMs) since the launch of ChatGPT in November 2022. However, small language models (SLMs) are gaining popularity, especially in settings where computational resources are limited or real-time processing is crucial. In 2024, nearly every major AI developer released compact, high-performing models.

Key Highlights

• A smaller model uses less memory and computing power. Their lower number of parameters – SLMs typically have from a few million to a few billion parameters – allows for shorter training times, thus saving energy. SLMs consume less inference energy than LLMs because they do not generate answers from scratch by recombining patterns from massive data sets. Small models creation and optimization techniques are driving stronger performance over time, resulting in lower energy consumption. The ability of SLMs to process AI requests at the edge avoids energy-intensive cloud data transfers, saving power.
• SLMs are not meant to replace LLMs but rather complement them. Companies will prioritize smaller models tailored to their industry-specific needs, and also to cut costs and emissions.​ If more and more consumers and enterprises adopt SLMs, they can help address the growing concern about AI’s carbon footprint.

• The AI industry has been dominated by LLMs since the launch of ChatGPT in November 2022. However, SLMs are gaining popularity, especially in settings where computational resources are limited or real-time processing is crucial.
• With their suitability for industry-specific applications, SLMs offer scalability across diverse environments. Furthermore, the environmental costs associated with training and operating LLMs are becoming increasingly difficult to ignore. Several leading generative AI (GenAI) providers, including Microsoft, Meta, and Google, released SLMs in 2024.

• Rapidly growing AI-related energy consumption poses sustainability challenges, sparking a crucial debate. Models are increasingly powerful and complex, resulting in considerable energy consumption and, consequently, a larger carbon footprint. SLMs require less computational power than LLMs, resulting in lower overall power consumption.
• This report provides an overview of the leading SLMs and identifies applications for which they are more applicable than LLMs. It also analyzes their environmental impact and provides a future market outlook.