Seoul-based startup FuriosaAI, founded in 2017 by ex-Samsung/AMD engineer June Paik, has made a breakthrough by winning LG AI Research as its first significant commercial client for its custom AI chip, RNGD (short for “Renegade”) Omni EkonomiThe Economic TimesTechCrunch.
LG’s Rigorous Seven-Month Review
After extensive benchmarking focused on both performance and energy efficiency, LG approved RNGD to power its ExaOne large-language model lineup 雅虎財經+10Omni Ekonomi+10Bloomberg.com+10.
Performance & Efficiency Gains
RNGD delivers an estimated 2.25× better LLM inference performance per watt compared to traditional GPUs GuruFocus+9GamesBeat+9The Economic Times+9.
It supports high-end LLMs like ExaOne 3.5 and 4.0, handling large context windows (4K and 32K tokens), delivering >60 tokens/sec with far less power usage GamesBeat+1TechCrunch+1.
A Vote of Confidence Over Meta
Months after declining an $800 million acquisition offer from Meta, FuriosaAI's RNGD grabbed its first enterprise deal—signifying strong belief in its independent vision FINVIZ+13TechRadar+13AInvest+13.
Strategic Global Expansion
Following LG, FuriosaAI aims to open doors in the U.S., Middle East, and Southeast Asia by H2 2025, while continuing R&D and prepping for a future IPO 富途新聞+3AInvest+3AInvest+3.
Challenging Nvidia’s Dominance
RNGD’s specialized architecture positions FuriosaAI as a viable alternative to Nvidia and other AI chip startups such as Groq, SambaNova, and Cerebras GamesBeat+3AInvest+3GuruFocus+3Omni Ekonomi+2The Economic Times+2富途新聞+2.
Efficiency at Scale
With improved performance-per-watt and a 4U rack system powering multiple GPUs, RNGD promises lower total cost of ownership for AI model deployment AInvestThe Economic Times+3GamesBeat+3TechRadar+3.
Sovereign AI Hardware
For companies like LG aiming to own their AI stack (hardware + software), RNGD presents an attractive end-to-end solution with straightforward integration and open APIs FINVIZ+14GamesBeat+14TechRadar+14.
For FuriosaAI: LG’s endorsement marks a pivotal validation. To sustain momentum, they'll need to ramp production, attract more enterprise clients, and prepare for global rollout.
Industry Impact: This marks a growing shift toward domain-specific AI accelerators over general-purpose GPUs. If proven across more models and sectors, RNGD could accelerate a multipolar AI-chip market.
For Nvidia & Peers: Expect more bids for efficiency-optimized compute. The battle to define LLM inferencing hardware is heating up—with power and cost as the new key battlegrounds.