What changed
AWS Marketplace can now meter software usage during long-lived SageMaker bidirectional streaming inference sessions. Providers periodically report consumption while endpoint infrastructure remains billed separately; Deepgram is the launch partner for speech-to-text and text-to-speech.
Why it matters
Real-time speech vendors can align software charges with actual streamed usage instead of an hourly software fee. Buyers still pay for provisioned endpoint infrastructure, including idle periods, so total economics remain a two-part calculation.
What to do
Before adopting a Marketplace streaming model, calculate software units and always-on endpoint cost separately, test interrupted-session accounting and compare the bill with direct API pricing at expected concurrency.
Caveat
The launch solves Marketplace software metering, not idle infrastructure cost or model-quality validation. Deepgram is the first named implementation, and AWS does not establish savings for every traffic pattern.