FriendliAI was founded in 2021 as a spin-out from research at Seoul National University, with the aim of commercialising AI research to make deploying and running large language models fast, affordable, and reliable. The company opened an office in San Francisco in 2026.
The company's core product is an AI inference platform purpose-built to deliver faster inference through custom GPU kernels, smart caching, continuous batching, speculative decoding, and parallel inference. The platform provides instant access to over 590,000 Hugging Face models and claims to achieve 2× or greater inference speedups compared to standard approaches. Technical work spans AI inference optimization, large language models, and GPU kernel development.
FriendliAI also offers enterprise-focused products, including Dedicated Endpoints for custom or fine-tuned models and container solutions, both backed by 99.99% uptime SLAs. The company operates in the AI research and enterprise AI verticals.
