FuriosaAI
FuriosaAI was founded in Seoul in 2017 by June Paik and fellow engineers from AMD, Qualcomm, and Samsung, with a thesis that custom silicon — not repurposed GPU stacks — is the right foundation for efficient AI inference. Its first chip, Warboy (2021, Samsung 14 nm), proved the architecture; its second, RNGD 'Renegade' (unveiled at Hot Chips 2024, built on TSMC), targets hyperscale LLM inference with a Tensor Contraction Processor architecture that benchmarks at 2.25× better performance per watt versus comparable GPU configurations. Renegade entered mass production in 2025 and landed LG AI Research — which adopted RNGD for its EXAONE platform — as a flagship enterprise customer. In February 2025 Meta was reported in acquisition talks; in March 2025 FuriosaAI declined the reported ~$800M offer over disagreements on post-acquisition strategy, choosing instead to grow as an independent company. In July 2025 it closed a $125M Series C bridge round (Korea Development Bank, Industrial Bank of Korea, Keistone Partners, PI Partners, Kakao Investment), crossing the unicorn threshold at a ~$735M valuation and bringing total disclosed funding to ~$246M. As of late 2025 the company was seeking a Series D of $300–500M with Morgan Stanley and Mirae Asset Securities as advisers, targeting a 2027–2028 IPO once revenue and margins have scaled. FuriosaAI plans to ship 20,000 RNGD units in 2026, up from a few hundred in 2025, and is already developing a third-generation chip.
FuriosaAI is the most credible non-GPU inference chip challenger to emerge outside the US: it rejected an $800M acquisition from Meta to bet on independent scale, landed a unicorn valuation on real silicon that benchmarks ahead of GPUs on LLM efficiency, and is targeting a public listing — making it a bellwether for whether purpose-built inference ASICs can carve durable market share from Nvidia.
furiosa.ai ↗·AI Accelerators·
South Korea·private 
