FriendliAI
FriendliAI builds and runs an AI inference platform for large language model serving, combining custom GPU kernels and batching techniques to speed up deployment across multiple clouds.
FriendliAI, founded in 2021 as a spin-out from Seoul National University, develops and operates one of the fastest AI inference platforms. The company focuses on deploying and running large language models as quickly, cheaply and reliably as possible. It was founded and is led by Byung-gon Chun, the inventor of Continuous Batching, and has its roots in academic research on AI systems optimisation.
The company's technical work spans generative AI inference optimisation, large language model serving, custom GPU kernels, continuous batching, speculative decoding, parallel inference, smart caching and multi-cloud infrastructure. The FriendliAI Inference Platform combines these techniques in a purpose-built inference stack, offering single-click deployment of more than 590,000 Hugging Face models and, by the company's measure, inference speeds of 2x or faster.
Alongside the platform, FriendliAI offers Dedicated Endpoints and the FriendliAI Container, both aimed at enterprise workloads running custom or fine-tuned models and backed by a 99.99% uptime service-level agreement. Customers named by the company include LG AI Research, SK Telecom, NextDay AI and Upstage.
FriendliAI is headquartered in Seoul, with an office in San Francisco that opened in 2026, and operates across multiple clouds including AWS, GCP and OCI. The company's stated mission is to make AI inference fast, affordable and reliable.