Deepgram Scales Voice AI with AWS Trainium

Deepgram teamed up with AWS to turbocharge its Nova-3 speech-to-text model by running it on AWS Trainium2 for batch inference, nailing serious throughput and cost wins. The deep dive covers porting to AWS Neuron, kernel tuning tricks, and building a data-parallel Rust runtime—perfect if you're wrestling with ML inference at scale.