The archive · Hardware & Devices · Strategic decision · 2017–2025
FuriosaAI turns down Meta's $800M, then signs LG as first chip customer
Korean AI-chip startup FuriosaAI rejected Meta's $800M offer, then landed LG AI Research as first major customer for its RNGD inference chip.
FuriosaAI
What the business is
Seoul-based fabless startup designing AI accelerators — first Warboy, then RNGD — purpose-built to run large language models more cheaply than Nvidia GPUs.
Starting capital：~KRW 242.3B (~US$175M) raised by 2023 across Series A (KRW 8B), Series B (KRW 82.1B) and Series C (up to KRW 200B); valuation rose from KRW 239.2B (2021) to ~KRW 800B (2023) per TopDaily.
How it started
Founded in Seoul in 2017 by June Paik, a former AMD GPU designer and Samsung Electronics engineer, to prove Korea could build world-class AI chips. It shipped Warboy for vision workloads (deployed at Kakao data centers), then RNGD, a dual-HBM3 accelerator built for LLM inference.
What happened
Meta approached FuriosaAI in early 2025; in March reports said the startup rejected an $800M (~KRW 1.2T) offer over business strategy and structure, not price. On 2025-07-21 FuriosaAI announced LG AI Research as its first major customer: RNGD will power LG's EXAONE platform, with FuriosaAI claiming 2.25x better inference performance than competing GPUs.
How it ended up
Independent and scaling: the LG partnership is one of the few public endorsements of an Nvidia rival by a major enterprise; the ~15-person team is now pursuing customers in the US, Middle East and Southeast Asia.
Background
FuriosaAI was founded in Seoul in 2017 by June Paik, an AMD GPU designer and Samsung Electronics engineer, to prove Korea could build world-class AI semiconductors. Its first chip, Warboy, targeted computer-vision workloads and was deployed at Kakao data centers; its second, RNGD, is a dual-HBM3 accelerator purpose-built for running large language models. By 2023 the company had raised roughly KRW 242.3 billion (about US$175 million) and was valued near KRW 800 billion, closing in on unicorn status.
In early 2025 Meta opened talks to buy the startup, reportedly offering $800 million. Korean and international press said FuriosaAI turned the offer down over disagreements about post-acquisition strategy and structure, not price. CEO June Paik chose to keep the company independent, betting that RNGD's cost and energy advantages would win customers against Nvidia and other AI chipmakers.
Three months later the bet began to pay off: on July 21, 2025, FuriosaAI announced that LG AI Research had adopted RNGD for its EXAONE platform after a seven-month evaluation, with FuriosaAI claiming 2.25x better inference performance than competing GPUs. The deal made LG one of the first major enterprises to publicly endorse an Nvidia rival, and the ~15-person team began chasing customers in the US, Middle East and Southeast Asia.
What has to be true
- A purpose-built inference chip can undercut general-purpose GPUs on total cost of ownership, which is what finally won LG over.
- Rejecting Meta's $800M offer was only rational because the company had a customer path: the LG deal landed three months later and made independence credible.
- Sovereign AI demand — Korea wanting its own model (EXAONE) on its own silicon — gave a domestic startup a beachhead against Nvidia.
- A tiny team (about 15 people) forced focus: the company concentrated on one workload, LLM inference, instead of trying to be a general GPU vendor.
What can be applied
Turning down a huge exit only works if you can convert independence into customers: FuriosaAI made the refusal credible by closing LG months later — the customer, not the rejection, validated the bet.
Aftermath
As of 2026-09-02 FuriosaAI remains independent with a Seoul headquarters and a Santa Clara office. RNGD won final approval from LG AI Research after a seven-month evaluation and is being integrated into the EXAONE ecosystem across electronics, finance, telecom and biotech. CEO June Paik expects the partnership to reach beyond Korea as LG's global customers adopt EXAONE, and the company is pursuing new deals in the US, Middle East and Southeast Asia. The open question is whether 15 people can scale fast enough to turn one flagship endorsement into a real business against Nvidia's ecosystem.
Sources
- AI chip startup FuriosaAI reportedly turns down $800M acquisition offer from Meta
- Instead of selling to Meta, AI chip startup FuriosaAI signed a huge customer
- Nvidia Rival FuriosaAI Lands LG as First Major Customer for AI Chip
- A 'challenge' for Nvidia... AI semiconductor market 'dark horse' emerges
spotted an error? The archive wants to know.
Your turn
You just read one. Describe what you are building, and see who is betting on the same thing.
Free account · 3 free questions · no card