Case Study: Cartesia achieves 90ms real-time voice latency with Together AI GPU infrastructure

Cartesia achieves ~90ms voice latency with Together AI

Cartesia, an enterprise voice AI company, faced significant infrastructure challenges in running their real-time, streaming voice models. Their requirements for ultra-low latency, a custom inference engine for their State Space Model (SSM) architecture, and PCI-compliant deployments were not met by standard hosted platforms. To support these needs, they turned to Together AI for their GPU infrastructure.

Together AI provided dedicated GPU clusters, giving Cartesia deep cluster-level access to deploy and tune their custom SSM inference engine. This solution enabled Cartesia to achieve a reported 90ms model latency for voice generation and serve millions of audio minutes daily. Together AI also supported their multi-node H100 training and provided the PCI-compliant infrastructure required for their enterprise deployments.

Cartesia