FriendliAI builds a high-performance AI inference platform that makes deploying and running large language models fast, affordable, and reliable, commercializing research from Seoul National University.
FriendliAI builds a high-performance AI inference platform designed to make deploying and running large language models fast, affordable, and reliable. Founded in 2021, the company was spun out of Seoul National University to commercialize academic research, and is led by AI researchers including Byung-gon Chun, the inventor of Continuous Batching. Its technical work spans AI inference optimization, GPU kernel engineering, continuous batching, speculative decoding, parallel inference, smart caching, and multi-cloud infrastructure.
The FriendliAI Inference Platform combines custom GPU kernels, smart caching, continuous batching, speculative decoding, and parallel inference to deliver 2x+ faster inference. It provides single-click deployment access to over 590,000 Hugging Face models. For enterprises running custom or fine-tuned models, the company offers Dedicated Endpoints and container-based deployment options, both backed by a 99.99% uptime SLA, across AWS, GCP, and OCI.
FriendliAI operates from Seoul, where it originated, and San Francisco, where it opened an office in 2026, serving customers in generative AI, enterprise AI, and telecommunications worldwide. The company's work sits at the boundary of research and production engineering: systems-level problems in inference performance at scale, tackled with the rigour of the academic breakthroughs it was founded to commercialize.
