FuriosaAI raised $620M at $2.9B: the AI inference chip race has another player with real production
Direct answer
South Korea's FuriosaAI announced on October 28 and 29, 2026 the close of its $620 million Series D (KRW 820 billion), at a post-money valuation of about $2.9 billion (KRW 3.8 trillion) and a pre-money valuation of $2.28 billion, per the company release and HPCwire, TokenPost and Bloomberg coverage. The largest investor was Korea's National Growth Fund, with $280 million; international investors contributed $110 million, with Infinitum the largest of that group. The round quadruples the prior $735 million valuation reported by Bloomberg.
What FuriosaAI does and where the money goes
- Inference, not training: FuriosaAI designs processors optimized for inference, the stage where a trained model generates answers. That is where the recurring cost of every business running agents lives.
- RNGD in mass production since January 2026: its second-generation accelerator ships as a 180-watt air-cooled card and in server configurations with multiple accelerators.
- Production targets: the company targets 20,000 RNGD accelerators in 2026 and more than 40,000 in 2027, with memory and factory capacity secured, per the company.
- Named customers: Samsung SDS, LG Group and Upstage appear among RNGD customers and partners, with a Samsung SDS NPU-as-a-service and Upstage's Daum search summary cited as deployments.
- Third generation with Broadcom: the capital funds more RNGD production, a stronger software ecosystem and a third accelerator in development with Broadcom, a 2-nanometer multi-die design with HBM4 memory.
What it means for your business even if you never buy a chip
- Cost per answer falls when supply grows: every new inference provider pressures the price you pay per million tokens and per agent in production. Hardware diversification is one force behind the model price war.
- Energy efficiency is direct cost: a 180-watt air-cooled card changes the economics of serving models in owned data centers and in the cloud. Ask your provider what 1,000 answers cost, not just the model price.
- Public capital is in the game too: nearly half the round comes from state-directed Korean funds, per Science Monitor's analysis, which also notes the absence of direct investment from chip giants on the shareholder list. The valuation is funded ambition, not verified sales.
- Verify deployment before betting: mass production and named customers are stronger signals than a round. The company, founded in 2017, reaches this point with product in customers' hands, not just a prototype.
Note: this article summarizes FuriosaAI's October 28 and 29, 2026 release and HPCwire, TokenPost, TechStartups and Bloomberg via Apex News coverage. Production figures are company targets and the valuation is post-round, not an independent measurement of sales.
Want AI that works for your business?
Message us on WhatsApp and let's talk about automation and AI agents for your business.
WhatsApp inicIARelated articles
- AI spending nears $400B but only 23% scale agents
- Claude Haiku 5.5 costs up to 90% less: the price war continues
- 5 AI trends for business in 2026 and 2027
Frequently asked questions
How much did FuriosaAI raise and at what valuation?
$620 million in Series D, at a post-money valuation of about $2.9 billion and pre-money of $2.28 billion, announced October 28 and 29, 2026.
Who put in the most money?
Korea's National Growth Fund, with $280 million. International investors added $110 million, with Infinitum the largest of that group.
What is RNGD?
FuriosaAI's second-generation inference accelerator, in mass production since January 2026, offered as a 180-watt air-cooled card and in multi-accelerator servers.
Why does inference matter to a business?
Because it is the recurring cost of running agents and models in production: every answer your business serves is billed in compute. More hardware supply pressures that cost down.
