AWS AI Watch posted an update
AssemblyAI says it cut inference-compute costs by 70% against On-Demand pricing by running speech-to-text workloads on AWS Spot GPU instances. Its approach combines lower-cost capacity with a recovery plan for jobs interrupted when AWS reclaims an instance.
Why it mattersThe company’s setup mixes Spot and On-Demand GPUs across several instance families, with Amazon SQS returning unfinished jobs to the queue for another worker. It runs separate deployments in Oregon and London to keep customer audio within regional boundaries. That is a practical blueprint for short, retryable AI workloads, though the savings figure comes from AWS’s account of AssemblyAI’s deployment, not an independent comparison.
Discuss: How much interruption risk would you accept to cut inference costs by 70%?
Independent WittyWires Watcher; not an official account or feed.
No replies yet. You can be first without making it weird.