FuriosaAI raises $620 million to take on Nvidia with chips that sip power

FuriosaAI announced a $620 million Series D on Wednesday, Oct. 28, 2026, at a $2.9 billion post-money valuation, to scale production of its RNGD inference accelerator and build its next-generation chip.

Everyone wants to be "the Nvidia alternative," and most of them are still showing slides. Furiosa's argument is that it's actually shipping: a chip in volume production, two big Korean customers, and a pitch built around watts rather than raw speed — which is the thing data centers are actually running out of. Twenty thousand cards is a rounding error next to Nvidia's volumes, and the state-backed National Growth Fund writing the biggest check says as much about Korea's chip ambitions as about Furiosa's numbers. The real test is the 2nm Gen 3 with Broadcom, which won't show up for a while.

FuriosaAI announced a $620 million Series D on Oct. 28, 2026, at a post-money valuation of $2.9 billion. A Series D is a later round of private funding, named with a letter. Post-money valuation is the price of the whole company after the new money is counted in. Bloomberg reported the round roughly quadruples that valuation, about four times the last price. The round size and the $2.9 billion figure are FuriosaAI's, from its blog post the same day. The four-times comparison is Bloomberg's.

International investors in the round are Cisco Investments, Infinitum, Jericho Capital, NeoFort Holding, QUDA Capital, and Worldline Capital. Those names are FuriosaAI's.

Korea's National Growth Fund committed KRW 370 billion, about $280 million. KRW is the Korean won. The Export-Import Bank of Korea, UAMCO, and Hanwha Asset Management joined as new investors. Seed investors NAVER, via NAVER Ventures, DSC Investment, and KDB Capital invested again, along with KDB, Nextrans, and Kakao Investment. Those lines are FuriosaAI's.

RNGD, Furiosa's inference accelerator, reached mass production in January 2026. An inference accelerator is a chip that runs an AI model already trained, rather than teaching a new one. Furiosa says it will make 20,000 cards this year and has secured the memory and the fab capacity to more than double that in 2027. A fab is the factory that makes the chips. Those lines are FuriosaAI's.

RNGD is a 180-watt, air-cooled PCIe card. PCIe is the standard slot a card plugs into inside a server. Eight of the cards make the NXT RNGD Server: 4U, 3 kW, 4 petaFLOPS (FP8), and 384 GB of HBM3. 4U is a rack height, four standard units, so the box fits a cabinet a company already owns. A kilowatt is a measure of power. 3 kW is the server's power draw. A petaFLOP is a speed score for math. FP8 is a compact number format used for AI. HBM3 is fast stacked memory on the chip. Furiosa says the server is built for the racks and power budgets enterprises already have, with no new cooling. It claims RNGD matches RTX PRO 6000 throughput at the same card count and serves twice as many users per kilowatt. The RTX PRO 6000 is an Nvidia graphics card used for AI work. Those specs and that comparison are Furiosa's.

Samsung SDS has launched an NPU-as-a-Service offering powered by RNGD. An NPU is a neural processing unit, a chip built for AI math, and the service means renting that chip instead of buying the machine. LG Group companies are deploying RNGD in on-premises hybrid cloud environments. On-premises means the computers stay in the company's own building. A hybrid cloud mixes those with rented machines. Those lines are Furiosa's.

The money goes to scaling RNGD production and deployments around the world, and to a third-generation chip already in development with Broadcom and other partners. That chip uses a 2nm multi-die design with HBM4 and HBM4E memory. 2nm is the manufacturing generation, a way of saying how tightly the circuits are packed. Multi-die means several chips in one package. HBM4 and HBM4E are the next versions of that stacked memory. Those lines are Furiosa's.

Furiosa was founded in 2017 in Seoul. It is hiring in Seoul, Silicon Valley, Lisbon, and Singapore. Those lines are Furiosa's.

John Yetimoglu, chief investment officer at Infinitum, said, "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." That quotation is his, on Furiosa's post.

The picture is FuriosaAI's NXT RNGD Server opened up to show its eight red RNGD inference cards. The credit is FuriosaAI.

In plain terms, FuriosaAI on Wednesday said it raised $620 million at a $2.9 billion valuation. Bloomberg says that is about four times the last price. The company is already building its RNGD card, 20,000 of them this year, and selling a server of eight cards that plugs into racks companies already have. Samsung SDS and LG are using it. The new money also pays for a 2nm chip with Broadcom that is still in development.