FuriosaAI closes $620 million Series D for AI inference chips
Designs AI chips and the software needed to run trained models in data centers. Inference means using a trained model to generate an answer, rather than training it from scratch. RNGD entered mass production in January 2026; Samsung SDS and LG Group deployments disclosed. 180-watt, air-cooled inference card; Tensor Contraction Processor reduces data movement. 20,000 cards in 2026 and 40,000-plus in 2027 are production plans, not delivered totals. Scale RNGD production and global deployments. Develop third-generation silicon with Broadcom and other partners. Expand software, compiler and model support.
DTT summary based on FuriosaAI's report. The full article is on the publisher's site.
Additional reporting
$620 million closed Series D at approximately $2.9 billion post-money valuation, announced October 28, 2026.
National Growth Fund of Korea invested about $280 million as the largest investor. International participants include Infinitum, Jericho Capital, QUDA Capital, Cisco Investments, NeoFort Holding and Worldline Capital. Other participants include Export-Import Bank of Korea, UAMCO, Hanwha Asset Management, NAVER Ventures, DSC Investment, KDB Capital, Korea Development Bank, Nextrans and Kakao Investment.
RNGD accelerator cards and NXT servers for AI inference, the work of running a trained model to generate answers. The company develops silicon, compilers and deployment software together.
The Tensor Contraction Processor makes higher-dimensional tensor operations the basic unit of hardware and software work. In plain terms, it plans more of the model together to reduce data shuffling, a source of wasted energy.
RNGD is a 180-watt air-cooled PCIe card. Eight cards form a 4U server with 384 GB of HBM3, fast memory placed close to the processors.
RNGD entered mass production in January 2026. The company identifies Samsung SDS commercial NPU-as-a-service, LG Group on-premises deployments and Upstage live search summarization as customer uses.
FuriosaAI expects 20,000 RNGD accelerators in 2026 and more than 40,000 in 2027. These are company production plans, not verified shipment totals.
Scale RNGD production and international deployments, expand software and develop Gen 3 with Broadcom and other partners. Gen 3 targets a 2nm multi-die design with HBM4/4E memory; that is a roadmap, not the current product.
$125 million Series C bridge announced in July 2025. Participants included Korea Development Bank, Industrial Bank of Korea, Keistone Partners, PI Partners and Kakao Investment.
