On Tuesday, South Korean AI chip startup FuriosaAI announced a deal to supply its RNGD chip to corporations utilizing LG AI Research’s new EXAONE platform.
LG’s next-generation hybrid AI model EXAONE 4.0 was released last week, and RNGD is tuned for LLMs.
Electronics, finance, telecommunications, and biotechnology are targeted for various applications through collaboration.
FuriosaAI rejected Meta’s $800 million purchase bid three months ago to stay independent.
According to local media, post-acquisition business strategy and organizational structure disputes stalled the purchase, not price.
Meta wants to buy AI chipmakers like FuriosaAI to minimize its dependence on Nvidia.
FuriosaAI CEO June Paik told TechCrunch: “We want to continue our mission, and I think it’s an exciting opportunity at the same time,” when asked why the Meta deal fell through.
It significantly influences the firm and me to make AI computing more sustainable.”
With Meta M&A off the table, Paik wouldn’t say if the business is seeking new capital.
Paik says this new relationship will open business opportunities beyond South Korea.
“South Korea’s leading sovereign AI model is LG AI’s EXAONE. It won’t be limited to LG. This will be a key AI model in the Korean AI ecosystem.
We anticipate high demand for this EXAONE and our semiconductor solutions in South Korea and elsewhere. LG partners with and serves global customers.
Paik added that we expect those clients, particularly worldwide customers, to use this.
Paik further noted that LG AI’s adoption of Furiosa’s AI chip and accelerator is one of a big company’s few public endorsements of a Nvidia competitor. The startup won because their hardware is cheaper.
“We had to prove that our solution delivers strong performance and lowers total cost of ownership,” Paik added.
FuriosaAI says its RNGD accelerator surpassed LG AI Research’s EXAONE models by 2.25 times in inference performance. LG found FuriosaAI hardware more energy efficient, says Paik.
Furiosa’s GPU is AI-only. We support several AI models efficiently. Unlike GPUs, which are general-purpose processors, our architecture is designed for AI computation. We do not build our chip for rendering or mining, Paik remarked.
The Seoul-based startup with a Santa Clara headquarters employs 15 employees worldwide.
