FuriosaAI raises $620 million at a $2.9 billion valuation to scale its RNGD inference chip

FuriosaAI, a Seoul-based maker of AI inference chips, announced a $620 million Series D on October 29 at a post-money valuation of about $2.9 billion. Dealroom reports that the valuation is roughly four times the company's previous $735 million mark and that US firm Infinitum Partners led the round. South Korea's National Growth Fund was the largest investor at $280 million, existing shareholders put in $125 million and international investors $110 million, per Pulse 2.0. Other new backers include Jericho Capital, QUDA Capital, Cisco Investments, Worldline Capital and Abu Dhabi's NeoFort Holding, and the Export-Import Bank of Korea made its first direct investment in a venture company. Digital Today reports the round in won as 820 billion at a valuation of about 3.8 trillion, enlarged on investor demand from an original target of 450 billion to 750 billion won.

The chip and its customers

The money goes to increasing production of RNGD, FuriosaAI's inference accelerator, which entered mass production in January 2026, along with expanding its software ecosystem and developing a third-generation accelerator with Broadcom. The company describes RNGD as a 180W air-cooled card with throughput comparable to GPUs that use more than twice the power, which is its own claim. Dealroom reports TSMC will produce about 20,000 RNGD chips this year and more than 40,000 in 2027, and that the Broadcom chip is not expected to sample until 2028.

Named customers are Korean: LG runs its EXAONE models on RNGD, Upstage uses it for search summarization, and Samsung SDS has launched an NPU-as-a-service offering built on it. Dealroom notes FuriosaAI turned down an $800 million acquisition offer from Meta in 2025. "We are entering a new phase of bringing our technology to the global AI infrastructure market at scale," CEO June Paik said.

Why it matters

Inference, the cost of running a model each time someone uses it, is where most business AI spending ends up, and it is the part of the market where specialized chips have the best chance against Nvidia. A $620 million round backed heavily by the Korean state gives one of those challengers the money to ship in volume, though tens of thousands of cards a year is still a small supply. Few companies will buy these chips directly. The effect for most buyers arrives through cloud and service providers, such as Samsung SDS's service, offering cheaper inference capacity. It is worth asking your providers which hardware sits behind their pricing, since the per-token rate depends on it.