FuriosaAI, a South Korean artificial intelligence chipmaker, has entered into a significant partnership to supply its RNGD accelerator to enterprises using LG AI Research’s newly launched EXAONE 4.0 platform.

The RNGD chip is tailored for large language model (LLM) performance. LG unveiled EXAONE 4.0, a next-generation hybrid AI model, last week. The collaboration targets sectors including electronics, finance, telecommunications and biotechnology.

This development follows FuriosaAI’s decision to reject an $800 million acquisition offer from Meta approximately three months ago. According to local media reports, the deal collapsed due to disagreements over post-acquisition strategy and organisational structure, rather than valuation.

Meta’s interest in acquiring FuriosaAI aligns with its broader objective to reduce dependence on third-party suppliers, such as Nvidia.

When asked why the Meta deal failed, FuriosaAI chief executive June Paik told TechCrunch: ‘We want to continue our mission. I believe it is an exciting opportunity, and an impactful contribution, both personally and for the company, to make AI computing more sustainable.’

While Paik declined to confirm whether FuriosaAI is currently seeking fresh investment, he emphasised that the LG partnership may unlock international opportunities.

"EXAONE is viewed as South Korea’s leading sovereign AI model. It will not remain exclusive to LG. We expect substantial demand across the Korean AI ecosystem and from global partners," he said.

The decision by LG AI Research to adopt FuriosaAI’s chip marks a rare public endorsement of a non-Nvidia AI solution by a major technology firm. According to Paik, one decisive factor was the hardware’s lower total cost of ownership.

‘We needed to prove that our solution offers not only high performance but also cost efficiency,’ Paik said.

FuriosaAI claims its RNGD accelerator delivered 2.25 times the inference performance of comparable GPUs when tested with EXAONE. LG also reported greater energy efficiency using the chip.

Unlike general-purpose GPUs, FuriosaAI’s chip is purpose-built for AI applications. ‘We support a wide range of AI models efficiently. Our architecture is designed specifically for AI, not for rendering or cryptocurrency mining,’ Paik said.

Headquartered in Seoul with an office in Santa Clara, California, FuriosaAI employs a global team of just 15 people.