AIINEWS

AIINEWS

Global intelligence on artificial intelligence and computing

AWS adds usage-based billing for bidirectional AI streaming on SageMaker

AIINEWS Editorial Board•technology
15 November, 2026•2 min read
AWS adds usage-based billing for bidirectional AI streaming on SageMaker

Executive Briefing

“AWS has announced a way for model providers to report usage during bidirectional AI...”

Key Takeaways

  • 1.AWS says model providers can report actual usage during streaming sessions for AWS Marketplace software charges. SageMaker AI endpoint infrastructure charges still apply separately.
  • 2.The mechanism adds a WebSocket connection for metering alongside the inference connection. The model server sends usage records through that companion connection while the session runs.
  • 3.Deepgram is the launch partner, bringing usage-based billing for speech-to-text and text-to-speech models to AWS Marketplace.
  • 4.The billable unit depends on the provider’s counting rules and product offer. AWS gives the example of 720 units for 12 minutes of billable audio in a product priced per audio second, but a connection open for 12 minutes does not by itself determine the billable audio quantity.
  • 5.By default, inference continues if the metering connection fails, although metering stops. Providers can instead stop inference on connection failure, but that setting does not guarantee billing for malformed or dropped records.

AWS has announced a way for model providers to report usage during bidirectional AI streaming sessions on SageMaker for software billing through AWS Marketplace. Customers still pay separately for SageMaker AI endpoint infrastructure.

Key Developments and Technical Context

AWS says model providers can report actual usage during streaming sessions for AWS Marketplace software charges. SageMaker AI endpoint infrastructure charges still apply separately.

The mechanism adds a WebSocket connection for metering alongside the inference connection. The model server sends usage records through that companion connection while the session runs.

Deepgram is the launch partner, bringing usage-based billing for speech-to-text and text-to-speech models to AWS Marketplace.

The billable unit depends on the provider’s counting rules and product offer. AWS gives the example of 720 units for 12 minutes of billable audio in a product priced per audio second, but a connection open for 12 minutes does not by itself determine the billable audio quantity.

By default, inference continues if the metering connection fails, although metering stops. Providers can instead stop inference on connection failure, but that setting does not guarantee billing for malformed or dropped records.

Strategic Analysis and Industry Implications

For products priced by audio processed, reporting records throughout a session could align software charges more closely with consumption than hourly software pricing, which can include idle periods. It does not remove separate infrastructure charges.

Providers need to reconcile server records with AWS Marketplace reports and buyer charges, especially for interrupted sessions. Sending a record over the connection does not establish that the usage was billed.

Outlook and Key Factors to Watch

The supplied information does not give specific Deepgram product prices or the total cost of a given workload once infrastructure charges are included.

The supplied information provides no measured results showing how reported usage matches actual charges for interrupted sessions.

Content and editorial synthesis copyright by AIINEWS Editorial Board
Share this intelligence report:
Expanded Perspectives & Intelligence

Related Dispatches & Deep Dives

Curated 8 specialized reports

Explore AIINEWS Intelligence Vault

Over 2,000 in-depth dispatches structured across 18 specialized tech & AI categories.

Browse all 18 categories