AWS case study: Exowatt cuts AI costs up to 49x with SageMaker Tuning

AWS published a case study saying Exowatt used Amazon SageMaker Tuning to build two task-specific models that beat frontier models on accuracy while cutting costs by up to 49x. AWS also warned that evaluation bias can lead teams to overestimate large general models without clean ground-truth tests.

• Exowatt built two task-specific models with SageMaker Tuning
• Inference costs were up to 49x lower than larger models
• Serverless tuning avoided reserving dedicated infrastructure
• AWS flagged evaluation bias without clean ground-truth tests