Introduction
On July 21, 2025, South Korea–based FuriosaAI announced a major deal with LG AI Research: supplying its next-gen RNGD AI inference chips to power LG’s EXAONE LLM platform. This marks the startup’s first public enterprise win, achieved after evaluating performance superiorities compared to GPU alternatives.
Background
Launched in 2020 by ex-Samsung and KAIST engineers, FuriosaAI developed RNGD (or “Renegade”) — an inference-optimized chip promising 2.25× better inference-per-watt and 30% energy savings versus GPUs.
Details of the Partnership
LG tested RNGD across EXAONE’s LLM workloads over seven months. According to Lead Product Unit Head Kijeong Jeon, RNGD achieved significantly lower power consumption and total cost of ownership—boosting LG’s push for sovereign AI infrastructure.
Strategic Decision
Three months earlier, FuriosaAI turned down an $800 million acquisition offer from Meta, citing a strategic priority to maintain independence and global expansion.
Expert Commentary
June Paik, CEO of FuriosaAI, noted: “Partnering with LG validates our architecture’s power and cost advantages for real-world AI systems.” Analysts view this win as significant for breaking Nvidia’s factory-like grip over AI inference hardware in Asia.
Market Impacts
- For FuriosaAI: Solid enterprise credentials and future agreements expected from industries in robotics, healthcare, cloud.
- For LG: Access to proprietary AI acceleration that supports national tech sovereignty goals.
- For the Ecosystem: Accelerates innovation among niche AI chip startups, reducing lock-in to general GPUs.
Challenges
- Scaling TSMC 5nm chip production efficiently.
- Software stack robustness and broad LLM compatibility.
- Scaling sales team and OEM outreach internationally.
Future Outlook
Initial RNGD shipments to LG IT labs are expected in late 2025. Long-term potential includes co-branded hardware commercial rollout under LG’s EXAONE then wider licensing deals.
