Deepgram ports Nova-3 speech model to AWS Trainium2
How Deepgram optimized voice AI at scale with AWS Trainium
The engineering detail on kernel tuning and the Rust runtime shows how a third-party speech vendor squeezed performance out of AWS's in-house silicon, a signal for teams weighing GPU vs. Trainium for inference cost. Practitioners running large-scale batch transcription can borrow the porting and tuning patterns directly. Compared with running Nova-3 on GPUs, the results offer a concrete cost baseline for AWS-native stacks.
Deepgram worked with AWS Forward Deployed Engineering to port its Nova-3 speech-to-text model to AWS Trainium2 for offline batch inference. The write-up details porting to the AWS Neuron stack, kernel-level tuning, and a data-parallel Rust runtime. It reports the throughput and cost results achieved for batch transcription workloads on Trainium2.
How Deepgram optimized voice AI at scale with AWS Trainium
Deepgram worked with the AWS Forward Deployed Engineering team to port its Nova-3 speech-to-text model to AWS Trainium2 for offline batch inference. This post covers porting to the AWS Neuron stack, kernel-level tuning, a data-parallel Rust runtime, and the throughput and cost results on batch transcription.
