SettledHardware
Deepgram ports Nova-3 to AWS Trainium2, reports up to 70% more audio per dollar
Deepgram and AWS say Deepgram's Nova-3 speech-to-text model now runs on Trainium2 for offline batch transcription, built with TorchNeuron, custom NKI kernels and a Rust runtime. Deepgram reports 11 to 70 percent more audio per dollar than its previous compute. The figures are vendor-reported, relative, and cover batch workloads only.
- HEAT
- OUTLETS
- 1
- FIRST SEEN
- LAST SIGNAL
Primary source
AWSaws.amazon.com/blogs/machine-learning/how-deepgram-optimized-voice-ai-at-scale-with-aws-trainium/Coverage · 2 articles, oldest first
Monday, October 19, 2026