Tokenta Signal SIGN IN TERMINAL >
2026.11.07 13:37 UTC
● ISSUE 256 · MICROSOFT × ORACLE × AI CLOUD · Reuters report of Bloomberg News (Duke FM syndication)● ISSUE 255 · TERAWULF × Q3 2026 × WULF · TeraWulf Q3 2026 results (GlobeNewswire release)● ISSUE 254 · PSI × QUANTUM · PSI Aaronson-Ambainis proof announcement (PSI)● ISSUE 253 · IBM × QUANTUM · IBM Quantum System Two Chicago unveiling (PR Newswire)● ISSUE 252 · OKLO × Q3 2026 · Oklo Q3 2026 Form 10-Q (SEC)● ISSUE 251 · VISTRA × Q3 2026 × VST · Vistra Q3 2026 results (PR Newswire)
● ISSUE 256 · MICROSOFT × ORACLE × AI CLOUD · Reuters report of Bloomberg News (Duke FM syndication)● ISSUE 255 · TERAWULF × Q3 2026 × WULF · TeraWulf Q3 2026 results (GlobeNewswire release)● ISSUE 254 · PSI × QUANTUM · PSI Aaronson-Ambainis proof announcement (PSI)● ISSUE 253 · IBM × QUANTUM · IBM Quantum System Two Chicago unveiling (PR Newswire)● ISSUE 252 · OKLO × Q3 2026 · Oklo Q3 2026 Form 10-Q (SEC)● ISSUE 251 · VISTRA × Q3 2026 × VST · Vistra Q3 2026 results (PR Newswire)
MARKETS & INDUSTRY · SIGNAL REPORT

After Rejecting an $800 Million Buyout, the Chip Valuation Quadrupled

South Korean inference chip company FuriosaAI closed a $620 million Series D at a $2.9 billion valuation, about four times the previous round, with the National Growth Fund contributing $280 million; RNGD is in mass production, with capacity set to double in 2027.

ISSUE 97 · 2026-10-28 · SOURCE / FuriosaAI official announcement · STATUS / VERIFIED · TOKENTA SIGNAL RESEARCH
FURIOSAAI × RNGDREPORTED + ANALYSIS
IMAGE: FuriosaAI official announcement — FROM THE ORIGINAL REPORT ↗
$620M
SERIES D
The Series D amount; the $2.9 billion post-money valuation is about four times the previous round.
$280M
STATE FUND
The National Growth Fund's contribution, close to half the round's total.
20K→40K+
RNGD OUTPUT
Planned RNGD output: 20,000 units in 2026, over 40,000 in 2027.
KEY TAKEAWAYS

WHAT HAPPENEDWhat Happened

On October 28, South Korean inference chip company FuriosaAI announced it had closed a $620 million Series D at a $2.9 billion post-money valuation, roughly four times the previous round's $735 million. The round was led by U.S.-based Infinitum Partners, with South Korea's National Growth Fund contributing $280 million and international institutions including Cisco Investments taking part.

The background to the raise was written in March 2025: FuriosaAI rejected an acquisition offer of about $800 million from Meta. A year and a half later, the company raised at a $2.9 billion valuation, and the ledger of that refusal has a public number for the first time.

The product is the second-generation inference chip RNGD: a 180-watt air-cooled PCIe card that entered mass production in January 2026, with plans to produce 20,000 units this year and more than double to over 40,000 in 2027. Samsung SDS uses it to provide NPU cloud services, and LG Group uses it to run its EXAONE models.

THE STATE CAPITALA State Fund Leading Half the Round: the National-Team Moment in Chip Financing

The $280 million from South Korea's National Growth Fund is close to half the round's total. South Korea counts AI semiconductors as part of its national compute strategy: the fund steps in, and international institutions follow.

This is not purely financial investment: the state fund's backing is itself a form of credit guarantee, helping an inference chip company cross the trust threshold of "can it actually manufacture at scale."

THE INFERENCE BETThe Inference Chip Playbook: Skip Training, Compete on Power per Unit

FuriosaAI's RNGD uses tensor contraction as its basic compute unit rather than the 2D matrix processing of Nvidia GPUs. The company claims its design power is under half that of Nvidia hardware.

The logic of the inference market differs from training: customers are not buying peak compute but how many requests each watt and each dollar can run stably. A 180-watt air-cooled card targets precisely the layer of the data centre that is easiest to replace.

THE GEN 3 ROADMAPA Third-Generation Chip with Broadcom: the Next Hand Is Already Dealt

The stated uses of the funding are two: expand RNGD's global commercialisation, and develop a third-generation hyperscale AI infrastructure solution with Broadcom. Producing one generation while developing the next, on two parallel tracks.

The company is also reportedly planning an IPO as early as 2027. The Series D's real examination is not the valuation but whether the 40,000-unit capacity planned for 2027 can be delivered on schedule.

THE SHIFTWhat Changed

FROM EXIT TO OUTPUT
From selling the company to selling capacity
One generation in production, the next in development; every use of the funding points at delivery.
FROM TRAINING TO INFERENCE
From training compute to inference power
A 180-watt air-cooled card targets the data-centre layer that is easiest to replace.

SIGNAL JUDGMENT — ANALYSISAfter Rejecting $800 Million, the Valuation Quadrupled

Rejecting an acquisition is not pride; it is a bet on manufacturing capability. Tokenta Signal's judgment is unchanged: the inference chip war will be decided not at launches but in 2027's delivery numbers.

WHAT REMAINS UNCERTAINWhat Remains Uncertain

The boundaries of this report: the funding figure follows the company's official announcement of $620 million (Korean media converted it to about $623 million from won; the difference is exchange-rate methodology). The claim of energy consumption "under half of Nvidia's" is the company's own and has not been measured by a third party. This does not constitute investment advice.

EVIDENCEEvidence

Key points confirmed from the original report: FuriosaAI's official announcement of a $620 million Series D at a $2.9 billion post-money valuation; the National Growth Fund contributed $280 million and Infinitum led the round; the second-generation RNGD inference chip entered mass production in January 2026, with output planned at 20,000 units in 2026 and over 40,000 in 2027; a third-generation chip is being developed with Broadcom.

ORIGINAL: FuriosaAI closes $620 million Series D to scale efficient, production-ready AI inference ↗
furiosa.ai · 2026-10-28 · FuriosaAI official announcement

PRIMARY SOURCESPrimary Sources

RELATEDRelated Intelligence

CORRECTIONS: if a fact on this page is wrong, we keep a correction record on the page. Contact team@tokenta.space