aipopdb

9 Nov 2026  ·  6,605 works · 976 news indexed

News / FuriosaAI raises $620 million Series D at a valuation of about $2.9 billion

FuriosaAI raises $620 million Series D at a valuation of about $2.9 billion

AIPopDB Newsroom · via jonpeddie.com · · 4 min read · South Korea

Read the original ↗
Image for FuriosaAI raises $620 million Series D at a valuation of about $2.9 billion

Image: Jon Peddie Research

FuriosaAI, a chip designer based in Seoul, announced on 29 October 2026 that it has closed a $620 million Series D at a post-money valuation of about $2.9 billion. That is roughly four times the $735 million valuation set by its $125 million Series C bridge in July 2025, and Ventureburn describes it as the largest Series D ever raised by a South Korean chip company. The money is meant to scale production and deployment of its RNGD inference chip, build out the software around it and fund a third-generation chip being developed with Broadcom.

Korea's National Growth Fund, a KRW 150 trillion, five-year government programme for AI and chips, is the largest investor at about $280 million (KRW 370 billion). International investors added $110 million, a tranche led by the US firm Infinitum; some reports say Infinitum led the whole round, while Jon Peddie Research has the state fund anchoring it. Others include QUDA Capital of Singapore, Cisco Investments, NeoFort Holding of Abu Dhabi, Jericho Capital and Worldline Capital. Existing shareholders put in $125 million, among them Naver, DSC Investment, KDB Capital, Kakao Investment and Nextrans. Korea Development Bank added $22 million and UAMCO $30 million, and Export-Import Bank of Korea and Hanwha Asset Management joined as new investors, the bank making its first direct startup investment. The named tranches add up to $515 million; the release does not say where the other $105 million came from.

“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”

John Yetimoglu, chief investment officer, Infinitum, to CRN Asia · crnasia.com ↗

RNGD entered mass production in January 2026, made by TSMC. FuriosaAI plans 20,000 cards for 2026 and says it has secured fab and memory capacity for more than 40,000 in 2027. The card is a 180-watt air-cooled PCIe board, and eight of them go into the 4U NXT RNGD Server, rated at 3 kW, 4 petaflops of FP8 and 384GB of HBM3. It runs on the company's Tensor Contraction Processor architecture, which treats tensor contraction as the basic unit of work, and supports PyTorch, Hugging Face models and a drop-in replacement for vLLM.

The company says RNGD matches the throughput of Nvidia's RTX Pro 6000 at the same card count while serving twice as many users per kilowatt. Jon Peddie Research notes that the release gives no model, batch size, precision or latency for the comparison and says to treat it as a vendor claim until independent tests appear. Customers include LG Group, which runs its EXAONE models on RNGD and measured 2.25 times the performance per watt against the GPU system it compared; Upstage, which uses it for live search summaries; and Samsung SDS, which launched an NPU-as-a-Service in July 2026. A software update, SDK 2026.4, raised tokens per watt across 18 models, by 138% on gpt-oss-120B.

The RNGD inference accelerator card shown without its cooler

The RNGD inference accelerator card shown without its cooler

Image: Ventureburn

The third-generation chip is to use 2nm, multi-die packaging and HBM4/4E memory, aimed at frontier agentic models in large clusters; no tape-out date has been given. FuriosaAI is hiring in Seoul, Silicon Valley, Lisbon and Singapore.

The round follows months of preparation. Bloomberg reported in January 2026 that the company had hired Morgan Stanley and Mirae Asset Securities, then aiming at $300 million to $500 million ahead of a listing, and Seoul Economic Daily reported that investor demand pushed the final figure above target. Meta reportedly offered about $800 million in early 2025, which FuriosaAI turned down. Its rival Rebellions raised $400 million pre-IPO in March 2026 at about $2.34 billion, and FuriosaAI is the second company backed under the government's "K-Nvidia" initiative.

About FuriosaAI

FuriosaAI is a Seoul-based fabless chip designer that makes accelerators for running AI models, with a stated aim of chips that run advanced models more sustainably. It was founded in 2017 by former Samsung engineers and is led by founder and CEO June Paik.

In 2021 it launched Warboy, a first-generation vision NPU on Samsung's 14nm process, and became the first AI chip startup to outperform Nvidia on MLPerf Inference; in 2022 it formed ecosystem partnerships with Hugging Face, Samsung, TSMC and LG AI Research. It unveiled the second-generation RNGD data-centre accelerator in August 2024, and by February 2025 had raised about $115 million in equity. Bloomberg reported in March 2025 that it turned down an $800 million takeover offer from Meta.

Jon Peddie Research says shipment counts, customer expansions and third-generation disclosures will show whether the valuation holds, and expects an IPO announcement in 2027, likely at the end of the second quarter, though that timing is its own estimate rather than a company statement; Ventureburn says a listing is reportedly eyed as early as 2027. Naver Ventures partner Yong Jung Park said Naver can add the most value by helping the company in North America.

Second figure from Jon Peddie Research's walk-through of the RNGD software stack and card

Second figure from Jon Peddie Research's walk-through of the RNGD software stack and card

Image: Jon Peddie Research