> ## Content Index
> Fetch the complete content index at: https://venture-block.com/llms.txt
> Use this file to discover other available public pages before exploring further.

# FuriosaAI raises $620M at a $2.9B valuation to double its inference chip output
- URL: https://venture-block.com/furiosaai-620m-series-d-rngd-inference-chips/
- Published: 2026-10-30T07:06:31.000Z
- Updated: 2026-10-30T07:06:31.000Z
- Description: The Seoul chip designer's Series D, anchored by Korea's National Growth Fund, quadruples its valuation and will fund more RNGD inference cards and a 2nm third-generation chip developed with Broadcom.
- Author: Venture Block
- Tags: Funding, Science

South Korean chip designer FuriosaAI has closed a [$620 million (KRW 820 billion) Series D](https://www.businesswire.com/news/home/20261028655033/en/FuriosaAI-Secures-%24620-Million-Series-D-to-Meet-Surging-Enterprise-AI-Inference-Demand-and-Accelerate-Next-Gen-Silicon?ref=venture-block.com) at a post-money valuation of about $2.9 billion, and says it will use the money to more than double production of its RNGD inference accelerator next year and to fund a third-generation chip it is developing with Broadcom.

The valuation is roughly four times the $735 million set in the company's previous round last year, [according to Bloomberg reporting cited by The Next Web](https://thenextweb.com/news/furiosaai-series-d-620m-2-9bn-valuation?ref=venture-block.com). The round also came in well above the original plan. FuriosaAI began raising in January with a target of KRW 450 billion to 750 billion, [Yonhap reported](https://www.yna.co.kr/view/AKR20261029044100017?ref=venture-block.com), and Bloomberg said at the time that the company had hired [Morgan Stanley and Mirae Asset Securities](https://www.bloomberg.com/news/articles/2026-01-19/ai-chip-startup-furiosaai-plans-500-million-round-before-ipo?ref=venture-block.com) to advise on a round it described as coming before an initial public offering.

## Who wrote the checks

The largest single investor is South Korea's state-backed National Growth Fund, which committed KRW 370 billion, or about $280 million, [according to the company](https://furiosa.ai/blog/furiosaai-620-million-series-d?ref=venture-block.com). International investors put in $110 million in total. US-based Infinitum was the biggest of them, joined by Jericho Capital, QUDA Capital, Cisco Investments, NeoFort Holding and Worldline Capital.

Existing shareholders added $125 million, including an additional $22 million from Korea Development Bank. All three of FuriosaAI's original seed investors, NAVER (through NAVER Ventures), DSC Investment and KDB Capital, invested again, alongside Nextrans and Kakao Investment. New Korean backers include UAMCO, Hanwha Asset Management and the Export-Import Bank of Korea, which picked FuriosaAI as the first direct venture investment in its history. The export bank has since made a second such bet, [leading a $200 million Series C push for ammonia power startup Amogy](https://venture-block.com/amogy-kexim-200m-series-c-ammonia-power/).

FuriosaAI told [MoneyToday](https://www.mt.co.kr/en/future/2026/10/29/2026102908523397390?ref=venture-block.com) that the KRW 820 billion round is the largest investment ever secured by a Korean startup.

"Taking an advanced AI chip from design to volume production is an exceptionally difficult undertaking, and few companies have demonstrated the ability to do so successfully," Infinitum chief investment officer John Yetimoglu said in the company's announcement.

## What RNGD does

RNGD (pronounced "renegade") is built only for inference, the work of running trained models in production, rather than for training them. It entered mass production in January 2026\. Each card is a 180-watt, air-cooled PCIe board. Eight of them fill FuriosaAI's NXT RNGD Server, a 4U box that draws 3 kilowatts and carries 384 GB of HBM3 memory for a combined 4 petaFLOPS at FP8 precision, according to the company. That power envelope is the pitch: the server is designed to fit into the racks and cooling that enterprise data centers already have, without the liquid cooling retrofits that the newest GPU systems often need.

FuriosaAI says RNGD matches the throughput of Nvidia's RTX PRO 6000 at the same card count and serves twice as many users per kilowatt. Those are the company's own figures. It targets workloads running at up to about 100 tokens per second per user, which it describes as the range that matters most for enterprise deployments of open models such as Gemma 4 and Qwen, the same class of [open-weight models](https://venture-block.com/moonshot-kimi-k3-1-pro-open-weights/) that companies increasingly host themselves.

Underneath is what the company calls a Tensor Contraction Processor. GPUs and most other accelerators break model operations into fixed-size, two-dimensional matrix multiplications. FuriosaAI's design treats tensor contraction, a higher-dimensional operation, as the basic unit of work for both the silicon and the compiler, which lets the compiler plan data movement across a whole model and cut energy spent shuttling data around the chip. On software, the company says its SDK 2026.4 release this month raised tokens generated per watt across 18 models compared with the previous version, including a 138% gain on gpt-oss-120B.

## Customers, capacity and the next chip

FuriosaAI expects to produce 20,000 RNGD cards in 2026 and says it has secured the memory and fab capacity to make more than 40,000 in 2027\. TSMC manufactures the chips, according to Bloomberg. Named deployments include Samsung SDS, which launched an NPU-as-a-service offering on RNGD in July, LG Group, which runs its EXAONE models on the cards, and Upstage, which uses them to serve search summaries.

The third-generation chip is aimed at a different market: large clusters running frontier agentic models at hyperscale. FuriosaAI says it will use a 2nm, multi-die design with HBM4 or HBM4E memory, and is being developed with Broadcom and other partners.

The round extends a bet on independence. In March 2025, FuriosaAI [reportedly turned down an $800 million acquisition offer from Meta](https://techcrunch.com/2025/03/24/ai-chip-startup-furiosaai-reportedly-turns-down-800m-acquisition-offer-from-meta/?ref=venture-block.com), with negotiations breaking down over post-deal strategy and structure rather than price. The company was founded in 2017 by June Paik, who previously worked at Samsung Electronics and AMD. Its fundraising lands as large buyers look for alternatives to Nvidia's data center chips, a dominance that regulators are also examining, as in the Justice Department's [questions about Nvidia's Hugging Face deal](https://venture-block.com/doj-nvidia-hugging-face-antitrust-inquiry/).

FuriosaAI says it is hiring in Seoul, Silicon Valley, Lisbon and Singapore, and that it expects to share more customer wins and hardware roadmap details soon.