The platform supports a diverse customer base with various deployment options, including dedicated endpoints, serverless APIs, and containers, all offered under a usage-based pricing model.

FriendliAI, founded in 2021 by Byung-Gon 'Gon' Chun out of Seoul National University and now based in Redwood City, California, has developed a proprietary inference engine that supports over 420,000 models from open source and custom sources like Hugging Face, enabling faster and more cost-effective AI model deployment.

The company's technology applies deep system and algorithmic optimizations, setting it apart from competitors such as Fireworks AI, and supports broad model coverage to facilitate enterprise AI deployment.

FriendliAI claims its inference engine delivers up to 90% GPU cost savings and boasts the fastest inference performance in the market, making AI deployment more efficient for a diverse range of customers.

Since its founding, FriendliAI has focused on making AI model inference faster, cheaper, and simpler, with a strategic move to Redwood City in late 2023 to expand its market presence.

The company's revenue model is usage-based, measured by GPU hours or tokens and images processed, serving clients from startups like Scatter Lab to large corporations such as LG Electronics.

In 2025, FriendliAI has experienced rapid growth in both usage and revenue, with projections indicating a six to sevenfold increase compared to 2024, driven by its market-leading inference technology.

To handle dynamic and uneven inference traffic, FriendliAI employs innovative techniques like 'continuous batching,' which dynamically manages requests to maintain high GPU utilization and operational efficiency.

The company recently raised $20 million in a seed extension round led by Capstone Partners, with additional investments from Sierra Ventures, Alumni Ventures, KDB Investment, and KB Securities, highlighting strong investor confidence.

Capstone’s partner Eun-gang Song emphasized FriendliAI’s technical innovation and performance advantages, positioning it as a key enterprise AI scaling partner.

With AI inference now accounting for 80-90% of GPU usage, the market for efficient inference solutions like FriendliAI's is rapidly growing and becoming increasingly critical in AI deployment.

Founded by a former Microsoft and Facebook researcher, FriendliAI aims to reduce operational costs and improve efficiency for organizations through its advanced inference technology.