FuriosaAI Raises $620M at $2.9B as Korea's State Fund Backs Its RNGD Inference Chip

FuriosaAI, the Seoul inference-chip maker, closed a $620 million (KRW 820 billion) Series D at a post-money valuation of about $2.9 billion, the company said in its October 28 announcement. South Korea’s state-backed National Growth Fund put in $280 million, the largest single check, per the Business Wire release.

The numbers

• $620M Series D at roughly $2.9B post-money.
• $280M from Korea’s National Growth Fund; $110M from international investors led by U.S.-based Infinitum, with Jericho Capital, QUDA Capital, Cisco Investments, NeoFort Holding and Worldline Capital.
• $125M in follow-on money from existing shareholders, including $22M more from Korea Development Bank.
• 20,000 RNGD cards planned for 2026 and 40,000+ in 2027, with memory and fab capacity secured, per the company.
• About 4x the $735 million valuation reported after its July 2025 bridge round, per Techwire Asia.

The chip: inference under 200 watts

RNGD ships as a 180-watt, air-cooled PCIe card. Eight of them make the NXT RNGD Server: 4U, 3 kW, 4 petaFLOPS FP8 and 384 GB of HBM3, built to fit existing enterprise racks and power budgets without a cooling retrofit. Furiosa says RNGD matches NVIDIA RTX PRO 6000 throughput at the same card count and serves twice as many users per kilowatt. It has been in mass production since January 2026.

Who is running it

Samsung SDS launched a commercial NPU-as-a-service on RNGD in July, LG Group runs EXAONE deployments on it, and Upstage uses it to serve live search summarization, according to the release. Software ships on a steady cadence: Furiosa says SDK 2026.4 raised tokens per watt across 18 models, including by 138% on gpt-oss-120B versus the prior release.

Where the money goes

The round funds RNGD production, the software stack, and a third-generation chip for frontier agentic models in large clusters. Gen 3 is in development with Broadcom and other partners on a 2nm multi-die design with HBM4/4E memory.

What we are watching

The release discloses no revenue, backlog or chip pricing, so the $2.9B mark rests on deployments and volume targets. The efficiency claims are the company’s own, with no named third-party benchmark. And 40,000 cards in 2027 is a rounding error next to hyperscaler GPU orders; the test is whether Gen 3 lands on schedule and moves Furiosa from Korean anchor customers to global ones.