AWS adds usage metering for SageMaker streaming model charges

AWS announced usage-based metering for SageMaker AI bidirectional streaming, enabling model providers to report consumption during live sessions for AWS Marketplace software charges. Deepgram is the launch partner for speech-to-text and text-to-speech billing. A second WebSocket connection carries usage records alongside the inference stream. Providers define the billable unit and report consumption as a session progresses, allowing software charges to follow usage rather than time a model is available. Endpoint infrastructure charges remain separate and continue while an endpoint is provisioned. By default, inference continues if the metering connection fails. Providers can instead require inference to stop, trading service availability for metering continuity; malformed records can still be dropped.