FuriosaAI Secures $620 Million to Challenge Nvidia in Inference Chips
Seoul-based chipmaker reaches $2.9 billion valuation to scale production of energy-efficient AI hardware.
1Series D Funding Milestone
Seoul-based FuriosaAI successfully closed a $620 million Series D funding round, reaching a $2.9 billion valuation. This represents a fourfold increase from its previous valuation of $735 million last year. The National Growth Fund of South Korea led the investment, contributing $280 million to support the company's global expansion efforts.
2Diverse Global Investor Base
International investors contributed $110 million, led by the US firm Infinitum. Additional participants included Cisco Investments, Jericho Capital, and Singapore’s QUDA Capital Partners. Notably, QUDA founder Charlie Foo, a former Nvidia executive, now serves as an advisor. Existing shareholders, including NAVER and DSC Investment, also provided $125 million in follow-on capital.
3Strategic Focus on Inference
Unlike companies focusing on model training, FuriosaAI specializes in inference chips designed to run trained AI models efficiently. Investors are increasingly seeking alternatives to Nvidia hardware that consume less energy. The company’s RNGD chip, which entered mass production in January, is positioned to address these specific data center power requirements.
4Production and Memory Constraints
TSMC is currently manufacturing 20,000 RNGD units for the company this year. FuriosaAI has secured factory and memory capacity to produce over 40,000 units by 2027. This expansion is critical, as industry experts like Micron anticipate that global memory shortages will continue to tighten significantly through the year 2028.
5Performance and Efficiency Claims
The RNGD card operates at 180 watts and utilizes air cooling, simplifying data center integration. FuriosaAI claims the hardware matches the throughput of Nvidia’s RTX PRO 6000 while serving twice as many users per kilowatt. While these performance metrics are significant, independent verification of these specific efficiency claims remains currently unavailable.
6Early Adopters and Partnerships
Major Korean enterprises are already integrating the technology into their workflows. Samsung SDS launched a cloud service using RNGD in July, while LG Group utilizes the chips for its EXAONE models. Additionally, the AI firm Upstage employs the hardware for search summaries, with ongoing discussions expanding into the Asia-Pacific and Middle East.
7Future Hardware Development Roadmap
The company plans to allocate the new capital toward software improvements and the development of its third-generation chip. This upcoming architecture will feature a 2nm, multi-die design utilizing HBM4 memory. FuriosaAI is collaborating with Broadcom and other strategic partners to ensure the successful design and volume production of this next-generation hardware.
8Market Context and Expansion
Operating offices in the United States, Singapore, and Lisbon, FuriosaAI is aggressively scaling its international footprint. In July, the company expanded into Europe through a partnership with Equinix in Lisbon. This funding round follows a broader trend of massive capital inflows into inference chipmakers, including Etched’s recent $21 billion valuation milestone.