As the insurance industry expands its use of generative artificial intelligence (GenAI), Reinsurance Group of America (RGA) highlights the importance of addressing the environmental impact of AI operations.
GenAI tools, while offering productivity and operational efficiencies, require significant computing resources, which in turn increase energy consumption and carbon emissions.
One of the primary methods to mitigate these effects is optimizing infrastructure. RGA notes that shifting GenAI workloads to hyperscale cloud providers with energy-efficient data centers can lower emissions.
Providers offer custom silicon and cooling systems to reduce energy use, and they support elastic scaling, allowing insurers to adjust computing power based on actual demand. This avoids the inefficiencies of idle servers. Additionally, insurers can select data center regions powered by higher percentages of renewable energy to reduce their carbon footprint.
RGA emphasizes that using managed cloud services helps avoid unnecessary hardware provisioning and reduces operational complexity. These services automatically adjust for performance and efficiency, further minimizing power use.
For example, deploying large language models through platforms allows insurers to focus on model outputs without managing the underlying infrastructure.
The need to curb emissions within its use is not without merit. GenAI is poised to significantly influence the reinsurance sector in 2025, offering advancements in risk assessment, operational efficiency, and market growth.
The global GenAI in insurance market is projected to grow from $729.1 million in 2024 to $1.51 billion in 2025, reflecting a compound annual growth rate (CAGR) of 39.8%.
In North America alone, the GenAI insurance market is expected to rise from $407.6 million in 2024 to $449.5 million in 2025, with a projected value of $7.27 billion by 2034, indicating a CAGR of 33.4%.
Data efficiency is another critical area. Insurers handle substantial volumes of data for underwriting, claims, and fraud detection. RGA notes that curating and preparing high-quality data, rather than increasing raw data volume, allows models to train more efficiently, requiring fewer cycles and less energy. Fewer training rounds reduce carbon emissions and shorten the time needed to reach reliable outputs.
Model size and design also influence energy use. RGA points out that large, general-purpose models may be unnecessarily resource-intensive for many insurance applications. Smaller or task-specific models often deliver similar results at lower energy costs.
Techniques such as quantization, which reduces numerical precision in model weights, can shrink model size and speed up processing. This results in lower energy requirements without significant accuracy trade-offs.
Another area of focus is inference optimization. RGA explains that minimizing the length of chatbot memory, for example, by limiting context windows or starting fresh sessions for each inquiry, reduces token processing. This helps lower computational load and energy consumption.
Prompt caching – reusing previous computation results for recurring prompts – also shows promise in improving efficiency, particularly when multiple users or systems generate similar queries.
End-user behavior also contributes to resource use. RGA recommends training users to interact with AI systems in ways that reduce token usage, such as avoiding task-switching in single chat sessions. These behaviors can improve response quality and reduce overhead.
Monitoring and governance are central to managing GenAI’s environmental impact. According to RGA, tracking usage metrics such as token consumption and department-level budgets enables insurers to detect inefficiencies and align AI use with sustainability goals.
Certain tools also allow insurers to link infrastructure decisions to carbon emissions, providing data for ESG reporting and internal accountability.
What are your thoughts on this story? Please feel free to share your comments below.
