South Korean chip developer FuriosaAI has closed a $620 million Series D round at a post-money valuation of $2.9 billion, the company announced on October 29, 2026. The round was led by international investors including Infinitum, Jericho Capital, and Cisco Investments, while South Korea’s National Growth Fund committed about $280 million.
FuriosaAI builds accelerators designed specifically for AI inference, the work of running trained models for real users. Its RNGD chip entered mass production in January 2026, and the company says it will produce 20,000 cards this year, with capacity secured to more than double that in 2027. The funding will scale RNGD production worldwide and build a third-generation chip with Broadcom and other partners, using a 2nm multi-die design.
The company’s argument is economic as much as technical. Training a model happens once, but inference runs billions of times, and power is now the binding constraint in data centers. RNGD is a 180-watt, air-cooled PCIe card; eight of them form a server delivering 4 petaFLOPS in FP8, designed to slot into existing racks without a cooling retrofit. FuriosaAI says RNGD matches the throughput of Nvidia’s RTX PRO 6000 at the same card count while serving twice as many users per kilowatt.
Commercial traction is already visible. Samsung SDS has launched an NPU-as-a-service offering powered by RNGD, and LG Group companies are deploying the chips in hybrid cloud environments.
The round marks one of the largest bets yet on an alternative to Nvidia’s dominance in AI hardware, where GPU pricing for agents is already under scrutiny. The challenge now is execution: turning volume production into volume deployments. If inference demand keeps climbing as models like ChatGPT and Gemini serve ever more users, efficient inference silicon could become one of the most valuable layers of the AI stack.
Source: FuriosaAI
