FuriosaAI, a South Korean startup specialising in AI chip design, is reportedly aiming to secure between $300m and $500m in a Series D funding round, as it prepares for an initial public offering (IPO).

This funding round has Morgan Stanley and Mirae Asset Securities as its co-advisers, according to a report in Bloomberg. It aims to bolster FuriosaAI’s efforts to compete with industry leader Nvidia.

The funds will support the mass production of its second-generation RNGD chip, expand its global business operations, and aid in the development of a third-generation chip. The startup is considering a public listing as early as 2027.

FuriosaAI expects the first mass-produced shipment of its RNGD AI chips from TSMC later this month.

The company, founded in 2017 by June Paik, who has experience with Samsung Electronics and Advanced Micro Devices (AMD), focuses on high-efficiency AI inference chips.

FuriosaAI claims that its RNGD chip provides 2.25 times the inference performance-per-watt of traditional graphics processing units (GPUs).

In a previous funding effort in July 2025, FuriosaAI raised $125m in a Series C round, increasing its total funding to $246m at that time. Participants in this round included Korea Development Bank, Industrial Bank of Korea, Keistone Partners, PI Partners, and Kakao Investment.

The Series C round’s objective was to meet growing demand for FuriosaAI’s flagship AI chip and accelerate its roadmap.

The company’s mission is to make AI computing more sustainable by addressing the inefficiencies of power-hungry graphic processing units (GPUs). FuriosaAI claims that its technology offers a more efficient alternative for AI compute scalability.

In July 2025, LG AI Research adopted the RNGD chip for its EXAONE foundation models, reporting a 2.25 times better performance-per-watt compared to legacy GPUs.

The RNGD chip’s performance stems from its Tensor Contraction Processor (TCP) architecture, co-designed with a software stack specifically tailored for tensor contraction, which is essential to deep learning computations. This approach provides efficient custom dataflow paths for computations, reducing unnecessary data movement and resulting in lower costs and more flexible deployments.

In March 2025, FuriosaAI reportedly declined an $800m acquisition offer from Meta. According to a report in South Korean publication Maeil Business Newspaper, the negotiations broke down due to disagreements over post-acquisition business strategy and organisational structure, rather than issues related to the offer’s price.

In September 2025, FuriosaAI launched the NXT RNGD Server, a system that integrates its RNGD accelerators into a turnkey solution designed for efficient AI inference deployment in existing data centres.

The NXT RNGD Server is engineered to provide high performance on key AI workloads while fitting seamlessly into current data centre setups.

It comes with the Furiosa SDK and Furiosa LLM runtime pre-installed, enabling immediate application activation. This system optimises standard PCIe interconnects, eliminating the need for proprietary fabrics or special infrastructure, making it a practical and cost-efficient option for enterprises to scale their AI operations within their existing facilities.