Deepgram ports Nova-3 speech model to AWS Trainium2
Understand the story
Deepgram partnered with AWS to port its Nova-3 speech-to-text model to run on AWS Trainium2 chips, targeting offline batch workloads such as contact center analytics and medical transcription. On the technical side, the Deepgram team adapted the model using TorchNeuron and tuned it with NKI kernels, ultimately running 8 model replicas on a single EC2 trn2.3xlarge instance. In terms of performance, according to the AWS Machine Learning Blog, in batch transcription scenarios, the amount of audio processed per dollar increased by 11%–70% compared to before. The report was published on October 20, 2026, and is the latest information on the matter; there are no follow-up reports on the event.
AI overview of reporting · Updated 2 h前
Reporting timeline
Explore different aspects of the story through its reporting.
- AWS Machine Learning BlogHow Deepgram optimized voice AI at scale with AWS Trainium
Deepgram partnered with AWS to port its Nova-3 speech-to-text model to AWS Trainium2 for offline batch workloads such as contact center analytics and medical transcription. The team tuned kernels using TorchNeuron and NKI, ultimately running 8 model replicas on a single EC2 trn2.3xlarge instance, improving audio processed per dollar by 11–70% in batch transcription scenarios.
Story heat trend
Not enough continuous observations to show a trend yet.