South Korean AI chip startup FuriosaAI has secured a significant partnership with LG AI Research, just months after turning down an $800 million acquisition offer from Meta. The collaboration will see FuriosaAI's RNGD accelerator powering LG's EXAONE large language models across enterprise applications in electronics, finance, telecommunications, and biotechnology.

The partnership represents a notable validation of alternatives to Nvidia's dominance in AI hardware. LG AI Research's rigorous testing found that RNGD delivered 2.25x better performance per watt compared to GPU-based solutions, while generating 3.75x more tokens within the same power constraints. The RNGD Server, featuring eight accelerators in a single air-cooled chassis, achieved impressive benchmarks including 60 tokens per second for EXAONE 3.5 32B models.

"We want to continue our mission to make AI computing more sustainable," CEO June Paik told TechCrunch, explaining why the 15-person startup declined Meta's offer despite no disagreement on price.

Built exclusively for AI workloads rather than general-purpose computing, RNGD's specialized architecture enables superior energy efficiency—a critical factor as organizations seek to scale AI deployments cost-effectively. The partnership positions both companies to expand beyond South Korea's market into global opportunities.