# FuriosaAI Raises $620 Million at a $2.9 Billion Valuation, With Korea's State Growth Fund Supplying Almost Half

Times of AI Desk · 2026-10-29 · Industry

[https://timesof.ai/2026/10/furiosaai-620m-series-d-2-9b-rngd](https://timesof.ai/2026/10/furiosaai-620m-series-d-2-9b-rngd)

> South Korean inference-chip maker FuriosaAI raised a $620 million Series D at a $2.9 billion post-money valuation, with Korea's National Growth Fund committing about $280 million. The company says it will produce 20,000 RNGD cards this year and has secured capacity to more than double that in 2027. Performance-per-watt claims against Nvidia hardware are the company's own.

FuriosaAI's new round is as much industrial policy as venture capital. The Seoul-based inference-chip maker said on **October 28** that it raised a **$620 million Series D** at a **$2.9 billion** post-money valuation, and Korea's **National Growth Fund** committed **KRW 370 billion**, about **$280 million**. By Times of AI's arithmetic, that is roughly 45% of the round.

## Who invested

Other backers include the Export-Import Bank of Korea, UAMCO, Hanwha Asset Management, NAVER Ventures, DSC Investment, KDB Capital, Korea Development Bank, Nextrans and Kakao Investment, the company says. International investors include Infinitum, Jericho Capital, QUDA Capital, Cisco Investments, NeoFort Holding and Worldline Capital; SiliconANGLE puts their combined share at about $110 million.

## What it is selling

FuriosaAI's **RNGD** inference card has been in mass production since January 2026. The company says it will produce 20,000 cards this year and has secured manufacturing capacity to more than double that in 2027. RNGD is a 180-watt PCIe card; FuriosaAI's NXT RNGD Server packs eight of them into a 4U, 3-kilowatt box with 4 petaflops of FP8 compute and 384GB of HBM3, according to the company.

FuriosaAI says RNGD matches the throughput of Nvidia's RTX PRO 6000 while serving twice as many users per kilowatt. Those are company figures; no independent benchmark was published with the round. Named customers include Samsung SDS, which offers an NPU-as-a-service built on RNGD, and LG Group companies running it in on-premises hybrid clouds, the company says. Its third-generation chip, in development with Broadcom and other partners, is planned as a 2-nanometre multi-die design with HBM4 or HBM4E memory.

"We are entering a new phase of bringing our technology to the global AI infrastructure market at scale," co-founder and chief executive **June Paik** said, according to SiliconANGLE. The company's case rests on power: it argues that data centres will be limited by the cost and availability of electricity, so tokens per watt matter most. Hyperscalers are making the same argument for their own silicon, as in [Microsoft's Maia 300 plans](https://timesof.ai/2026/10/microsoft-maia-300-inference-chip-2027). FuriosaAI's pitch is that customers who cannot build their own chip can still buy efficiency outside Nvidia.

## Limits

- Throughput and users-per-kilowatt comparisons are FuriosaAI's own; no third-party benchmark accompanied the announcement.
- The 45% state share is Times of AI's calculation from the company's KRW and dollar figures; exchange rates move.
- The 2027 capacity figure describes secured supply, not orders.
- The third-generation chip has no published ship date.

## Sources

- [FuriosaAI: FuriosaAI raises $620 million Series D (October 28, 2026)](https://furiosa.ai/blog/furiosaai-620-million-series-d)
- [SiliconANGLE: Energy-efficient inference chip developer FuriosaAI raises $620M to solve the data center power bottleneck (October 28, 2026)](https://siliconangle.com/2026/10/28/energy-efficient-inference-chip-developer-furiosaai-raises-620m-to-solve-the-data-center-power-bottleneck/)