Case Study: Hedra achieves 60% cost savings and 3x faster inference with Together AI

Hedra cuts video inference costs 60% with Together AI

Hedra, an AI-native startup specializing in viral video generation, faced the challenge of scaling its infrastructure to handle unpredictable viral demand without incurring ballooning costs or requiring dedicated platform engineers. Their custom model architectures and need for elastic performance were not met by traditional hyperscalers. They partnered with Together AI, using its Dedicated Container Inference service, to overcome these hurdles.

Together AI provided a solution featuring joint engineering collaboration for kernel optimizations, a flexible technical architecture, and dynamic auto-scaling. This partnership resulted in a 60% reduction in total costs, 3x faster inference on Blackwell hardware, and the ability to support 300x growth in GPU usage. These savings and performance gains enabled Hedra to maintain healthy margins while serving millions of videos per month.

Hedra