Recommended Articles
Fundraising

FuriosaAI Raises $620 Million Series D at $2.9 Billion Valuation to Challenge Nvidia in AI Inference

By Angeled · October 30, 2026 · Fundraising
FuriosaAI Raises $620 Million Series D at $2.9 Billion Valuation to Challenge Nvidia in AI Inference
FuriosaAI Raises $620 Million Series D at $2.9 Billion Valuation to Challenge Nvidia in AI Inference

Seoul-based FuriosaAI has completed a $620 million Series D financing at a $2.9 billion post-money valuation, one of the largest AI chip rounds announced in the week of October 28, 2026. The deal, reported by Bloomberg on October 28 and confirmed by the company the same night, roughly quadruples the company's previous valuation of about $735 million. It positions a specialist inference-chip maker as a serious, if still unproven, challenger in a market dominated by Nvidia.

What was announced

FuriosaAI said the Series D was completed on October 28, 2026. South Korea's National Growth Fund contributed about $280 million, or roughly 45 percent of the round. International investors contributed about $110 million, according to Bloomberg, with the balance coming from Korean financial institutions and existing shareholders.

The international backers named include Infinitum Partners, Jericho Capital, QUDA Capital of Singapore, Cisco Investments, NeoFort Holding of Abu Dhabi, and Worldline Capital. Returning investors include Naver, Korea Development Bank, DSC Investment, Nextrans, and Kakao Investment. The Export-Import Bank of Korea also made what was described as its first direct investment in a startup.

What FuriosaAI actually builds

FuriosaAI was founded in 2017 and designs processors optimised for AI inference — the stage at which a trained model generates answers, rather than the training stage at which it learns. Inference is where the recurring cost of AI lives: every chatbot reply, agent action, and generated image consumes compute again.

Its second-generation processor, RNGD, entered mass production in January 2026. The company says RNGD is available as a 180-watt accelerator card and in server configurations holding multiple accelerators. It expects Taiwan Semiconductor Manufacturing Company (TSMC) to produce about 20,000 RNGD cards in 2026, and says it has secured memory and fabrication capacity to more than double that output in 2027.

FuriosaAI has pointed to deployments involving Samsung SDS and LG Group, and it is developing a third-generation accelerator with partners including Broadcom. The Broadcom chip is not expected to sample until 2028, which makes the next two years a test of execution: turning a mass-produced second-generation chip into repeat, production-scale orders.

Why inference, and why now

Training a frontier model grabs headlines, but inference decides whether an AI product is commercially viable. As applications serve millions of users, or as autonomous agents perform repeated tasks, power consumption, processor utilisation, and total cost of ownership compound quickly. A chip that is merely fast on a benchmark is not enough; it must be efficient under real workloads, compatible with the models customers actually run, and supported by reliable software.

That is FuriosaAI's bet: that processors designed specifically for inference can beat general-purpose graphics processors on efficiency for this workload. It is also its central risk. Nvidia's advantage is not only silicon. It is a software platform, a developer ecosystem, and deep relationships with cloud providers. A challenger must win on hardware, software, supply, and support at the same time.

Why it matters for founders and investors

First, inference economics are becoming a board-level question. If a startup's unit costs rise with every user, its margins shrink precisely as it scales. Founders choosing model providers, cloud infrastructure, and hardware should be able to explain their inference cost per thousand requests, not just their model's quality.

Second, the round shows how AI infrastructure funding is splitting. Some capital chases frontier model labs. A growing share chases the picks and shovels: chips, power, cooling, and data-centre capacity. FuriosaAI's $620 million sits squarely in that second bucket.

Third, the valuation jump — from about $735 million to $2.9 billion — illustrates how quickly hardware valuations can re-rate once a chip reaches mass production. Production, not a prototype, is what changed the story. For hardware founders, that is a practical fundraising lesson: the round that follows first silicon at scale looks very different from the round that funds the design.

What to watch next

Three questions will decide whether this round ages well. Can FuriosaAI convert its 2026 production run into large, repeatable customer orders beyond its named Korean deployments? Can its software stack keep pace with new models, so customers are not locked into a narrow set of workloads? And can it hold its production ramp into 2027, when it plans to more than double output? The company has disclosed no revenue figure in connection with this round, so investors and observers must judge progress by shipments, customer announcements, and follow-on orders rather than by the valuation alone.

Angeled.com connects founders with investors.

Bloomberg, reported October 28, 2026, via company confirmation and subsequent coverage
FuriosaAI company announcement, October 28, 2026
Dealroom News, October 28, 2026
TechStartups, October 29, 2026

Read Next