FuriosaAI Raises $620 Million Series D at $2.9 Billion Valuation
FuriosaAI has closed a $620 million Series D at a post-money valuation of $2.9 billion. The company says the money will go toward scaling production of its RNGD inference accelerator and developing a third-generation chip.
- FuriosaAI closed a $620 million Series D at a $2.9 billion post-money valuation.
- RNGD is in mass production, with 20,000 cards planned this year and capacity to more than double that in 2027, per the company.
- Samsung SDS and LG Group companies are among the named RNGD customers.
- The third-generation chip is in development with Broadcom, using a 2nm multi-die design and HBM4/4E memory.
- SDK 2026.4 raises token generation per watt across 18 models, including by 138% on gpt-oss-120B, according to FuriosaAI.
Investors and production plans
According to FuriosaAI, the round included Cisco Investments, Infinitum, Jericho Capital, NeoFort Holding, QUDA Capital and Worldline Capital. Korea's National Growth Fund committed KRW 370 billion (about $280 million). The Export-Import Bank of Korea, UAMCO and Hanwha Asset Management joined as new investors. All three original seed investors, NAVER (through NAVER Ventures), DSC Investment and KDB Capital, contributed again.
The company says its RNGD inference accelerator is in mass production. It plans to produce 20,000 cards this year and says it has secured the memory and fab capacity to more than double that in 2027. FuriosaAI says RNGD reached mass production in January 2026.
What RNGD offers
FuriosaAI frames its design around three needs: performance, efficiency and programmability. RNGD is a 180-watt, air-cooled PCIe card built to deliver the most tokens possible from less than 200 watts. Eight cards make up the NXT RNGD Server, which the company lists as 4U, 3 kW, 4 petaFLOPS (FP8) and 384 GB of HBM3. It is designed to fit the racks and power budgets enterprises already have, with no cooling retrofit.
The company claims RNGD matches RTX PRO 6000 throughput at the same card count and serves twice as many users per kW of power. It targets models running at up to about 100 tokens per second per user, and says it runs open models such as Gemma 4 and Qwen 3.8-27B well.
Underneath is what FuriosaAI calls the Tensor Contraction Processor (TCP). Where GPUs and most other accelerators break model operations into fixed-size, two-dimensional matrix multiplications, TCP makes tensor contraction, a higher-dimensional operation, the basic unit of work. The company says this lets its compiler plan across a whole model and, thanks to deterministic execution, minimize energy-wasting data movement.
Customers and software
FuriosaAI says it is deploying RNGD with AI cloud providers, sovereign AI initiatives, AI-native companies and enterprises building on-premises and hybrid cloud infrastructure. Samsung SDS has launched an NPU-as-a-Service offering powered by RNGD, and LG Group companies are deploying the card in on-premises hybrid cloud environments.
On software, the company says it has shipped four major SDK releases this year, adding hybrid batching, prefix caching and native Kubernetes support. SDK 2026.4 raises token generation per watt across 18 models compared with 2026.3, including by 138% on gpt-oss-120B, according to FuriosaAI. Customers can deploy in PyTorch as they would on GPUs, and write their own kernels in the Tensor Contraction Language and Virtual ISA when they need more control.
Gen 3 and what comes next
The round has two priorities: scaling RNGD production and deployments globally, and building a third-generation chip to serve frontier agentic models in large clusters. Development is underway with Broadcom and other partners. Gen 3 will use the TCP architecture with a 2nm, multi-die design and HBM4/4E memory, the company says.
FuriosaAI argues that the world will need far more inference compute and that power cost and availability will constrain every facility, from hyperscalers to small on-premises sites. John Yetimoglu, Chief Investment Officer with Infinitum, said the company "has gone further, securing production deployments with leading enterprises such as Samsung SDS and LG Group." For buyers, the pitch is an inference option that fits existing racks and power limits. The company is also hiring in Seoul, Silicon Valley, Lisbon and Singapore.
