FriendliAI, founded in Seoul in 2021, specialises in optimising how large language models are served. The company's core technology accelerates AI inference - the process of running trained models in production - through techniques including continuous batching, speculative decoding, custom GPU kernels, and parallel inference. Its platform promises more than twice the inference speed of standard approaches and provides instant access to over 590,000 models from Hugging Face.
The company offers three main products: a managed AI inference platform, dedicated endpoints for enterprise clients running custom or fine-tuned models, and container-based deployment solutions. Enterprise offerings come with 99.99% uptime SLAs. The firm serves customers across AI research, telecommunications, and the broader technology sector.
FriendliAI opened an office in San Francisco in 2026 alongside its original Seoul headquarters. Financial terms and funding details have not been publicly disclosed.
