# Deepgram cuts speech‑to‑text costs 11‑70% with AWS Trainium2

> By porting its Nova‑3 model to AWS’s Trainium2 chips, Deepgram processes more audio per dollar on batch jobs.

Oossa · 2026-10-19 · https://oossa.com/en/deepgram-cuts-speech-to-text-costs-11-70-with-aws-trainium2

Deepgram moved its Nova‑3 speech‑to‑text model to AWS Trainium2 chips. The port runs eight model replicas on a single trn2.3xlarge instance. In offline batch jobs like contact‑center analytics, Deepgram measured 11‑70 % more audio processed per dollar compared with its previous compute options. The work involved using TorchNeuron for PyTorch, swapping layers with custom kernels, and building a Rust runtime to avoid Python overhead.

## The facts

- Deepgram’s Trainium2 setup processed 11–70 % more audio per dollar.
- Eight Nova‑3 replicas run on a single trn2.3xlarge EC2 instance.

## Why it matters

Customers using Deepgram’s API can expect lower transcription costs for large‑scale batch workloads.

## Sources & references

1. [How Deepgram optimized voice AI at scale with AWS Trainium](https://aws.amazon.com/blogs/machine-learning/how-deepgram-optimized-voice-ai-at-scale-with-aws-trainium/) – AWS, 2026-10-19

Last updated: 2026-10-19