The AI race isn’t just about who can train the biggest model. That’s a billionaire hobby. The real question is who can serve those models at scale without setting fire to GPU budgets. FriendliAI just answered it with a $20M seed extension led by Capstone Partners, with Sierra Ventures, Alumni Ventures, KDB Capital, and KB Securities backing once again. Total funding now sits at $26.8M, and the move signals one thing: inference is the real battlefield.
FriendliAI was founded in December 2021 by Byung-Gon Chun and Gyeong-In Yu, two engineers who saw the GPT-3 explosion not as the finish line but as the spark for a new problem. Byung-Gon Chun, Ph.D. from UC Berkeley, cut his teeth at Microsoft, Yahoo!, and Intel Research before taking a professorship at Seoul National University. Gyeong-In Yu, Ph.D. from Seoul National University, earned the Best Dissertation Award before co-founding FriendliAI. Together, they realized the bottleneck wasn’t building giant models, it was delivering them to enterprises without melting hardware budgets.
Their answer is Friendli Engine, a generative AI inference platform built for speed and efficiency. Continuous batching, quantization down to FP8 and INT8, speculative decoding, Multi-LoRA serving, it’s a pit crew built for language models. Enterprises running on FriendliAI report up to 90% GPU cost savings while hitting the fastest inference benchmarks in the industry. Artificial Analysis ranked them no. 1. Integration with Hugging Face means over 430,000 models are one click away. Scatter Lab cut GPU costs in half. NextDay AI tripled throughput. SK Telecom, Upstage, DeepAI, TUNiB, Twelve Labs, the receipts keep stacking.
This raise isn’t a lifeline, it’s a growth accelerant. FriendliAI will expand sales and marketing to pull in more enterprise customers, scale engineering to push inference optimization further, and strengthen global partnerships that already stretch from Redwood City to Seoul. Securing exclusive API access to LG AI Research’s EXAONE 4.0 wasn’t a lucky break, it was proof FriendliAI is becoming core infrastructure in the generative AI economy.
The lesson here isn’t complicated. Training is flashy, but inference is where enterprises either scale or stall. FriendliAI is betting that efficiency, not excess, will decide who wins the next chapter of AI. With $20M in fresh capital and a platform that delivers speed without waste, the odds are stacked in their favor.
