Illustrative image: a high-performance accelerator card held in hand. FuriosaAI has not published a license-safe official product photo on a stable source, so this is an illustrative photo of the kind of inference accelerator RNGD is — not FuriosaAI's card. Image: Wikimedia Commons.
TL;DR: FuriosaAI Series D funding explained in one line: the Seoul AI chip startup has closed a $620 million Series D at a $2.9 billion valuation, led by South Korea's National Growth Fund, to scale production of its RNGD inference accelerator and build a next-generation chip with Broadcom — one of the largest AI hardware bets of 2026.
📌 KEY HIGHLIGHTS
- $620 million Series D: FuriosaAI closed the round (KRW 820 billion) at a post-money valuation of about $2.9 billion, announced 29 October 2026.
- State-backed lead: South Korea's National Growth Fund put in about $280 million (KRW 370 billion), the round's largest cheque; international investors added $110 million.
- RNGD in mass production: the second-generation inference accelerator entered mass production in January 2026; FuriosaAI plans 20,000 cards this year and more than 40,000 in 2027.
- Real customers, not demos: Samsung SDS runs its NPU-as-a-Service on RNGD, LG Group runs EXAONE deployments on it, and Upstage serves live search summarisation with it.
- Next up, Gen 3: a third-generation accelerator with Broadcom — 2nm, multi-die, HBM4/4E memory — aimed at large agentic inference clusters.
FuriosaAI Series D funding explained: what happened?
FuriosaAI, founded in Seoul in 2017, builds chips designed for one job only: AI inference — running trained models for millions of users, rather than training them. On 29 October 2026 the company announced the completion of its Series D, a $620 million (KRW 820 billion) round that values it at approximately $2.9 billion post-money. The round was upsized from an initial target of KRW 450–750 billion after demand from domestic and international investors, according to Korean press reports.
The valuation is nearly four times the roughly $735 million reported after its $125 million Series C bridge financing in July 2025 — a jump that reflects a real change in the company's status, from chip designer to volume manufacturer. Reports last year said FuriosaAI turned down an $800 million acquisition offer from Meta; the company chose to stay independent, and this round is the payoff for that bet.
Who put in the money?
The investor list mixes Korean state capital, global chip specialists and returning early backers. South Korea's National Growth Fund committed KRW 370 billion (about $280 million), making it the largest single investor — part of Seoul's push to build a domestic AI semiconductor industry. International investors contributed $110 million in total, led by US-based Infinitum, alongside Jericho Capital, Singapore's QUDA Capital, Cisco Investments, Abu Dhabi-based NeoFort Holding and Worldline Capital.
Existing shareholders added $125 million in follow-on funding, including all three original seed investors — NAVER (through NAVER Ventures), DSC Investment and KDB Capital. New Korean institutions joined too: the Export-Import Bank of Korea made its first direct investment in a startup, UAMCO committed $30 million, and Hanwha Asset Management joined the round.
| Round detail | Figure |
|---|---|
| Series D size | $620 million (KRW 820 billion) |
| Post-money valuation | About $2.9 billion (about KRW 3.8 trillion) |
| Largest investor | National Growth Fund of Korea — about $280 million (KRW 370 billion) |
| International investors | $110 million total — Infinitum, Jericho Capital, QUDA Capital, Cisco Investments, NeoFort Holding, Worldline Capital |
| Existing shareholders | $125 million follow-on, including NAVER, DSC Investment, KDB Capital, Nextrans, Kakao Investment |
| Previous valuation | About $735 million after the July 2025 Series C bridge |
| Use of funds | Expand RNGD production and global deployments, software ecosystem, third-generation chip with Broadcom |
What is RNGD, and why does efficiency win?
RNGD (pronounced "renegade") is FuriosaAI's second-generation accelerator, built on its proprietary Tensor Contraction Processor (TCP) architecture. Instead of breaking AI work into fixed two-dimensional matrix multiplications like most accelerators, TCP makes tensor contraction — a higher-dimensional operation — the basic unit of work for both hardware and software. The practical claim: more tokens per watt, with a compiler that plans across a whole model and deterministic execution that cuts wasted data movement.
The card itself is a 180-watt, air-cooled PCIe card — deliberately modest on power so it slots into data centres' existing racks and cooling, with no retrofit. Eight cards make the NXT RNGD Server: 4U, 3 kW, 4 petaFLOPS (FP8), 384 GB of HBM3 memory. FuriosaAI says RNGD matches an RTX PRO 6000's throughput at the same card count while serving twice as many users per kilowatt — a company claim, now backed by production deployments rather than benchmarks alone. RNGD entered mass production in January 2026 on TSMC's 5nm process, and FuriosaAI says it has secured the memory and fabrication capacity to more than double output next year.
Illustrative image: server racks in a data centre. RNGD cards are designed to fit existing data-centre power and cooling budgets without retrofits. Image: Wikimedia Commons / NOIRLab / NSF / AURA / T. Slovinský, CC BY 4.0.
Why it matters for World Tech readers — and what it means for India
AI's bottleneck has shifted. Training grabs headlines, but inference — running models billions of times a day — is where the electricity bill lands, and power is now the scarce resource in data centres everywhere. FuriosaAI's pitch is that the winner in inference is not the fastest chip at any cost, but the most tokens per watt inside power budgets that already exist. A $620 million round, led substantially by a sovereign fund, is also a geopolitical signal: South Korea is explicitly building a domestic alternative to Nvidia, and FuriosaAI is its second company picked under the country's "K-Nvidia" initiative after Rebellions, according to Korean press reports.
For India, be precise: FuriosaAI has announced no India office, India pricing or India deployment with this round, so any India impact is second-order for now. The relevance is real but indirect — India's data-centre build-out faces the same power constraints, Indian enterprises run the same open-weight models RNGD targets (such as Gemma and Qwen), and the funding wave in AI silicon is driving down inference costs for everyone. It is also a lesson for Indian hardware founders: FuriosaAI's valuation quadrupled only after it reached volume production and named customers — manufacturing execution, not just chip design, is what investors paid for. That is the same lesson as this week's other big startup story — our Foundry Robotics $100 million Series A explainer — and the same funding momentum as Serval's $150 million Series C.
Should you care?
✅ Verdict: Yes, if you follow AI infrastructure, chips or startup funding — a $620 million round at a $2.9 billion valuation, with production hardware and named enterprise customers, makes FuriosaAI the most credible Nvidia challenger outside the US and China. If you are an enterprise buyer, the practical signal is narrower: inference efficiency is now a procurement criterion, not a footnote. Otherwise there is nothing to buy or do today — but watch the Gen 3 chip with Broadcom, because that is where FuriosaAI either becomes a platform or stays a niche supplier.
FAQs
What is FuriosaAI?
FuriosaAI is a Seoul-based semiconductor company founded in 2017 that designs AI accelerators for data-centre inference. Its proprietary Tensor Contraction Processor architecture and full software stack aim to deliver more tokens per watt than general-purpose GPUs for enterprise AI workloads.
How much did FuriosaAI raise in its Series D?
FuriosaAI closed a $620 million (KRW 820 billion) Series D, announced 29 October 2026. South Korea's National Growth Fund contributed about $280 million, international investors $110 million, and existing shareholders $125 million in follow-on funding.
What is the RNGD chip?
RNGD is FuriosaAI's second-generation AI inference accelerator, in mass production since January 2026. It is a 180-watt, air-cooled PCIe card built on TSMC's 5nm process with HBM3 memory, designed to run large language models efficiently within existing data-centre power and cooling limits.
Who invested in FuriosaAI's Series D?
The largest investor was South Korea's National Growth Fund (about $280 million). International backers included Infinitum, Jericho Capital, QUDA Capital, Cisco Investments, NeoFort Holding and Worldline Capital, alongside returning investors NAVER, DSC Investment, KDB Capital, Nextrans and Kakao Investment.
What is FuriosaAI's valuation after the Series D?
About $2.9 billion post-money (about KRW 3.8 trillion) — nearly four times the roughly $735 million valuation reported after its July 2025 Series C bridge round.
Takeaway: FuriosaAI turned volume production into a $2.9 billion valuation; the next test is whether its Broadcom-built Gen 3 chip can scale from enterprise deployments to hyperscale clusters.
Follow World Tech on TechByCountry for tomorrow's update.
📸 Get tech in the last 24 hours on Instagram — follow @_techinlast24hrs for daily posts (9 AM–9 PM) and reels at 12:30 PM & 8:30 PM.
Sources
- FuriosaAI — FuriosaAI closes $620 million Series D (company announcement, 29 October 2026)
- Business Wire — FuriosaAI secures $620 million Series D (press release, 28 October 2026)
- HPCwire — FuriosaAI secures $620M Series D (29 October 2026)
- WOWTALE — Furiosa AI raises USD 623M in Series D (29 October 2026; KRW 820 billion total, pre-money valuation detail)
Comments
Post a Comment