Furiosa AI, a South Korean AI chip company, has raised $623 million (820 billion won) in a Series D round, it said on October 29. The round values it at about $2.89 billion post-money, or $2.28 billion pre-money. The raise beat its original target of $342 million to $570 million. The National Growth Fund was the largest investor with $281 million, followed by US-based Infinitum. Cisco Investments, QUDA Capital Partners, and existing shareholders including Korea Development Bank and NAVER also took part. Furiosa AI will use the money to expand production and sales of RNGD, its second-generation AI chip, which entered mass production in January. It will also speed up development of a third-generation chip with Broadcom, aimed at hyperscale AI infrastructure.

FuriosaAI's valuation jumps fourfold to $2.9 billion as it chases Nvidia. Founder June Paik turned down a roughly $800 million Meta buyout offer in October 2025 because Meta wanted the chips locked up for its own data centres. Now FuriosaAI's $620 million Series D, led by Infinitum Partners with $280 million from Korea's National Growth Fund, backs his plan to sell RNGD chips to anyone. Seoul-based FuriosaAI just quadrupled its valuation to $2.9 billion as investors hunt for efficient alternatives to Nvidia's dominant AI chips. FuriosaAI closed a $620 million Series D round this week, according to Bloomberg. That pushes its valuation to $2.9 billion, from roughly $735 million in its prior funding. That's a four-times jump in a single raise. The investor list reads like a bet on Korea's chip independence as much as on one startup. South Korea's National Growth Fund was the round's largest investor, contributing $280 million, while Infinitum Partners was the largest international investor. International backers signed on too: Jericho Capital, Worldline Capital, Singapore's QUDA Capital, Abu Dhabi's NeoFort Holding and Cisco Investments. Existing shareholders Naver, Kakao Investment and Korea Development Bank came back for more. FuriosaAI was founded in 2017 by June Paik, a veteran of Samsung Electronics and AMD. It makes neural processing units built specifically for AI inference: the work of actually running a trained model, rather than training it in the first place. That distinction matters more than it sounds. Training chips need raw horsepower. Inference chips need to serve thousands of concurrent requests cheaply, and that's a different design problem entirely - one Nvidia's GPUs were not originally built to solve. FuriosaAI's flagship product is a second-generation chip called RNGD (pronounced "Renegade"). It packs about 40 billion transistors on a 5-nanometer TSMC process, paired with SK Hynix's HBM3 memory. Each unit delivers roughly 512 teraflops of compute and about 1.5 terabytes per second of memory bandwidth. The company says RNGD handles up to 7.4 times more concurrent users than Nvidia's RTX Pro 6000 at the same power level. In testing with LG AI Research's EXAONE models, RNGD delivered 2.25 times better inference performance per watt than the GPU system tested. Mass production started in January. TSMC is scheduled to build 20,000 RNGD units this year, scaling past 40,000 in 2027. FuriosaAI is already working with Broadcom on its third-generation accelerator. Samsung SDS didn't wait for that scale-up. In July it launched what it calls Korea's first subscription-based NPU cloud service, running entirely on RNGD chips inside the Samsung Cloud Platform. That gives FuriosaAI a live commercial deployment well before its IPO ambitions take shape. FuriosaAI has installed its RNGD inference accelerators at Equinix's Lisbon data center, its first commercial deployment in Europe, while pursuing a third-generation chip with Broadcom built on HBM4 memory. The move lands as European chip startups scramble for funding far behind their US rivals in the race for Nvidia alternatives. - alternative to Nvidia chips Europe - FuriosaAI Renegade inference accelerators Lisbon Here's the real signal in this story, and it isn't the valuation number. In March 2025, FuriosaAI rejected an acquisition offer from Meta worth roughly $800 million and chose to remain independent. Reports at the time said the negotiations broke down over differences about FuriosaAI's direction and organizational structure after a takeover, rather than the price. That bet looks better by the month. AI-driven demand has kept advanced memory supply tight, with SK Hynix saying in October 2026 that its DRAM and HBM supply was effectively sold out through 2029 and Micron reporting that most of its 2027 capacity had already been sold. When the memory that every AI chip depends on is rationed this tightly, any credible alternative to Nvidia's GPU stack becomes worth paying up for - so long as it can actually get built and shipped. Investors are buying more than a chip design here. They're buying a position in a supply chain where the scarce resource is no longer compute: it's capacity itself. FuriosaAI is reportedly eyeing a public listing as early as 2027. Morgan Stanley and Mirae Asset Securities advised on its Series D fundraising. Will $2.9 billion hold up by then? That depends less on RNGD's benchmark numbers and more on whether Broadcom's next chip ships on time. It also depends on whether the rest of the world's data centres stay desperate enough for a Nvidia alternative to keep paying this kind of premium for one. Join the discussion. No replies yet. Start the discussion. Judith Murphy is a financial journalist and market analyst covering AI, technology stocks, and emerging market trends. She has contributed to multiple financial publications and brings a data-driven approach to her coverage of the technology sector and its impact on global markets.

FuriosaAI valuation 2026: $2.9B after $620M Series D. South Korea's National Growth Fund led the round for the Seoul-based inference-chip maker, lifting its valuation roughly 4x in 15 months as it pitches lower-power alternatives to Nvidia's GPUs. Founder & Editor-in-chief, Value Add VC - Angel investor, VC, family office, operator and founder · 3x founder (BrandYourself, Launch.it, SPOT) · 65+ investments 65+Investments 3xFounder $200M+Funds Tracked Quick Answer FuriosaAI's valuation hit $2.9 billion on October 28, 2026, after the Seoul-based chipmaker raised a $620 million Series D led by South Korea's National Growth Fund. That's roughly 4x the $735 million valuation it set 15 months earlier in its July 2025 Series C bridge round. FuriosaAI's valuation reached $2.9 billion after it raised $620 million in a Series D round led by South Korea's National Growth Fund, announced October 28, 2026 - roughly 4x the $735 million valuation it set 15 months earlier. The Seoul-based chipmaker isn't a paper valuation story. LG Group, Upstage and Samsung SDS already run paying workloads on its inference chips, and the new money is funding a scale-up to tens of thousands of units a year as FuriosaAI tries to pry data-center inference workloads away from Nvidia on power efficiency rather than raw performance. FuriosaAI valuation 2026: How a $620M round set a $2.9B price tag. FuriosaAI's valuation is $2.9 billion as of October 28, 2026, set by a $620 million Series D led by South Korea's National Growth Fund, which put in $280 million of the round as part of a government-backed push to build a domestic chip industry. That is up from $735 million in July 2025, a roughly fourfold increase in 15 months. up from $735M in Jul 2025 Valuation (Oct 28, 2026) led by National Growth Fund Series D raised National Growth Fund stake 40,000 units 2027 chip production target SiliconANGLE reported that beyond the National Growth Fund, Cisco Investments, Abu Dhabi-based NeoFort Holdings, QUDA Capital Partners, Jericho Capital and Worldline Capital added a combined $110 million, with existing backers Korea Development Bank, Kakao Investment and NAVER rounding out the round. NAVER Ventures has said it has participated in every FuriosaAI round dating back to the company's 2017 pre-seed, under co-founder and chief executive June Paik. Who backed FuriosaAI before the Series D. FuriosaAI's July 2025 Series C bridge, which set the $735 million base this round built on, drew roughly 40 institutions, Korean outlet Asia Economy Daily reported. The largest disclosed checks, in Korean won as originally reported, show a round built mostly from domestic development banks and venture funds rather than global chip investors: | Investor | Amount (July 2025 bridge) | | Korea Development Bank | 30.0B won | | Keystone Partners | 20.0B won | | PI Partners | 13.2B won | | Eugene-Growth Finance Fund | 11.6B won | | Industrial Bank of Korea | 10.0B won | | IBK Securities | 6.0B won | | IBK Venture Investment | 5.0B won | | Daesung Startup Investment | 5.0B won | | Kakao Investment | 3.0B won | How Value Add VC counted: largest disclosed checks in FuriosaAI's 170B won (~$125M) Series C bridge; amounts as reported in won; as of July 31, 2025. Source: Asia Economy Daily. Verdict put FuriosaAI's total disclosed funding at $246 million after that bridge. Add the new $620 million Series D, and FuriosaAI's disclosed total funding is roughly $866 million, by Value Add VC's count - the company has not published its own cumulative figure. What FuriosaAI's chips do, and the pitch against Nvidia. FuriosaAI's current chip, the Tensor Contraction Processor, is the second generation built on the company's RNGD architecture. Instead of Nvidia's two-dimensional matrix math, it uses tensor contraction as the basic unit of computation with deterministic execution, which FuriosaAI says is built to maximize tokens produced per watt of power consumed - SiliconANGLE reports the company claims the chips use less than half the power of Nvidia's GPUs on the same inference workload, though the article does not name a specific Nvidia model or an independently measured figure. The chips run PyTorch and vLLM, the software stacks most AI teams already use for inference, which lowers the cost of switching away from Nvidia hardware compared with a fully custom stack. FuriosaAI is also co-developing a third-generation chip with Broadcom Inc., though SiliconANGLE's report discloses no terms or timeline for that chip beyond it following the current TCP generation. The company is targeting production of 20,000 TCP chips in 2026 and 40,000 in 2027, scaling further after that. Power efficiency, not peak throughput, is the bet: as AI data centers increasingly run into electrical-grid and power-procurement limits rather than chip-supply limits, a chip that does more inference per watt becomes a cost argument independent of whether it ever matches Nvidia's GPUs on raw speed. Who's already paying for FuriosaAI's chips. FuriosaAI isn't pre-revenue. LG Group uses the chips to power its EXAONE language models, Upstage runs live search-summary features for millions of users on the silicon, and Samsung SDS launched a commercial neural-processing-unit cloud service built on FuriosaAI hardware in the summer of 2026, according to SiliconANGLE. Real paying customers on real products is the detail that separates FuriosaAI from a startup whose valuation rests purely on a sovereign fund's industrial-policy check. How FuriosaAI's $2.9B compares to Groq, tenstorrent and cerebras. FuriosaAI is one of several well-funded challengers trying to take inference workloads from Nvidia, and its $2.9 billion valuation sits in the middle of the pack as of their most recent disclosed rounds. Groq's $3.5B is actually a down round - half the $6.9B it hit at its peak 11 months earlier, after Nvidia struck a licensing deal with the company. See its Groq valuation breakdown. Track every inference and training chip startup's valuation on the AI Chip Startups dashboard, and see how the inference-chip category differs from the training-chip race in inference chips vs training chips. Where Value Add VC could be wrong. The reporting on FuriosaAI's own numbers isn't as clean as the headline $2.9 billion suggests: outlets covering the July 2025 bridge round put the resulting valuation anywhere from roughly $700 million to $770 million depending on the won-to-dollar conversion and whether pre-money or post-money terms are being quoted, which means the "4x in 15 months" framing carries real rounding risk even though every version of the math still shows a sharp step-up. On the business itself, a sovereign fund contributing nearly half of a $620 million round is a genuine signal of South Korea's industrial policy, not purely market demand - LG, Upstage and Samsung SDS being paying customers helps, but none has disclosed contract size, so the revenue behind this valuation remains unverified. The Broadcom-codeveloped third-generation chip and the 40,000-unit 2027 production target are both unproven commitments, and FuriosaAI's power-efficiency claim against Nvidia's GPUs comes from the company itself, with no independent benchmark cited in SiliconANGLE's report. Explore related dashboards. Interactive tools with live data on this topic Track FuriosaAI and every other AI chip challenger on the AI Chip Startups dashboard at Value Add VC. Latest from the pulse. Get VC data most people never see - free to subscribe Trace's notes on venture, AI and startups, a few times a week. Join 5,000+ subscribers. No spam. Frequently asked questions. What is FuriosaAI's valuation in 2026? How much has FuriosaAI raised in total? Who invested in FuriosaAI's $620M Series D? How does FuriosaAI compete with Nvidia? Who are FuriosaAI's customers? Related tools & dashboards. New in AI & technology. Explore 45+ free VC tools, dashboards, and recommended startup software.