FuriosaAI Raises $620 Million at $2.9 Billion Valuation to Scale Its RNGD Inference Chip
South Korea's National Growth Fund put in about $280 million. The money will fund RNGD production and a 2nm third-generation chip being developed with Broadcom.
Published
Last updated / Fact-checked on
Seoul-based AI chipmaker FuriosaAI announced on October 28, 2026 that it has closed a $620 million Series D at a $2.9 billion post-money valuation. Bloomberg reported that this quadruples the company's previous valuation. The money will scale production of the RNGD inference accelerator and fund a third-generation chip aimed at large AI clusters.
Image: ArticCynda / CC0, via Wikimedia Commons
Who invested
| Investor group | Contribution |
|---|---|
| South Korea's National Growth Fund | KRW 370 billion (about $280 million) |
| International investors | $110 million (Bloomberg) |
| Total round | $620 million |
The international investors were Cisco Investments, Infinitum, Jericho Capital, NeoFort Holding (Abu Dhabi), QUDA Capital (Singapore) and Worldline Capital. The Export-Import Bank of Korea, UAMCO and Hanwha Asset Management joined as new investors. All three original seed investors returned: NAVER (through NAVER Ventures), DSC Investment and KDB Capital. Korea Development Bank, Nextrans and Kakao Investment also invested.
What RNGD is
FuriosaAI was founded in 2017. RNGD, its second-generation chip, is built only for inference, meaning running trained models rather than training them. It entered mass production in January 2026.
- Card: 180-watt, air-cooled PCIe
- Server: the NXT RNGD Server holds eight cards in 4U at 3 kW, with 4 petaFLOPS (FP8) and 384 GB of HBM3
- Company claim: matches Nvidia RTX PRO 6000 throughput at the same card count while serving twice as many users per kilowatt (company figures, not independently benchmarked)
- Volume: 20,000 cards in 2026, with memory and fab capacity secured to more than double that in 2027
The company says the server is designed to fit the racks and power budgets that enterprises already have, without cooling retrofits. FuriosaAI names Samsung SDS, which launched an NPU-as-a-Service offering on RNGD, and LG Group companies, which run it in on-premises hybrid clouds, as production customers.
Gen 3 and Broadcom
The round has two stated priorities: scaling RNGD globally and building a third-generation chip for frontier agentic models running in large clusters. FuriosaAI announced a partnership with Broadcom in May 2026. It says Gen 3 is in development with Broadcom and other partners. The chip will use a 2nm multi-die design with HBM4/4E memory, built on the company's Tensor Contraction Processor (TCP) architecture. FuriosaAI has not announced a release date or pricing.
On software, FuriosaAI says it has shipped four major SDK releases in 2026. It says SDK 2026.4 improves tokens generated per watt across 18 models, including by 138% on gpt-oss-120B.
Why it matters
Investors are looking for more energy-efficient alternatives to the Nvidia chips that dominate AI data centers, Bloomberg reported. FuriosaAI's case rests on tokens per watt. That pitch matters most to operators whose capacity is limited by power and cooling, not by floor space.
What to watch
- whether independent benchmarks back up RNGD's efficiency claims
- whether FuriosaAI wins customers outside Korea; it says it is hiring in Seoul, Silicon Valley, Lisbon and Singapore
- Gen 3 sampling dates and Broadcom's role as details emerge
For more on AI hardware, see our Artificial Intelligence hub.
How we reported this
We checked the deal terms against Bloomberg's report and FuriosaAI's announcement, and we confirmed the Broadcom partnership through search. We cut figures that neither source supported, including a $125 million existing-investor tranche. We also removed embedded third-party images and tracking links. Performance figures are attributed to the company.
Sources
- Bloomberg (via Bloomberg Law): FuriosaAI Quadruples Valuation to $2.9 Billion in Nvidia Chase, Oct 28, 2026.
- FuriosaAI: FuriosaAI closes $620 million Series D, Oct 28, 2026.
- FuriosaAI: Partnership with Broadcom to build next-generation inference platform, May 2026.
DailyMech used AI tools to help research and draft this article; it was fact-checked by our editors.