What changed
Exowatt announced the use of Amazon SageMaker AI serverless model customization to develop task-specific models for edge and cloud deployments. These models achieve higher accuracy than frontier models at up to 49× lower cost. The models are part of a repeatable AI pipeline that transforms raw operational data into certified training sets, fine-tuned models, and evaluated deployment artifacts.
What this means in practice
Exowatt's energy management system (EMS) operates across a growing fleet of sites, each generating continuous alarms. The company needed an AI agent to autonomously monitor and triage these alarms. Using SageMaker AI, Exowatt fine-tuned the Qwen3-1.7B model for edge alarm triage and the Qwen3-14B model for cloud energy-market analysis. These models are deployed in environments with limited or no internet connectivity and operate without GPU requirements for edge deployment. The solution enables Exowatt to run the AI pipeline in days rather than months, significantly improving operational efficiency.
Limitations and availability
The models are currently deployed for specific use cases: edge alarm triage and cloud energy-market analysis. Exowatt plans to reuse this approach to build smaller, task-specific models across its fleet. The solution relies on AWS services such as Amazon S3, SageMaker AI, and Amazon Bedrock. The models are available through the SageMaker Model Registry and can be invoked privately from the Cloud EMS service. The workflow requires no instances, clusters, or endpoints to manage, and charges only for the time the fine-tuning job runs.
