FuriosaAI Lands Major Customer, Bypassing Meta Deal

FuriosaAI Partners with LG AI Research to Supply RNGD AI Chips

FuriosaAI, a South Korean artificial intelligence chip developer, recently announced a collaborative agreement to provide its RNGD AI chip to businesses utilizing LG AI Research’s newly launched EXAONE platform. This partnership was revealed on Tuesday.

RNGD and EXAONE 4.0: A Synergistic Combination

The RNGD chip is specifically engineered for the efficient operation of large language models (LLMs). Just last week, LG introduced its advanced next-generation hybrid AI model, EXAONE 4.0. This collaboration is strategically focused on serving vital industries, including electronics, finance, telecommunications, and biotechnology, with a diverse array of applications.

Independent Path After Acquisition Offer

This announcement follows approximately three months after FuriosaAI chose to remain independent, declining a reported $800 million acquisition offer from Meta. The decision wasn't based on financial considerations.

According to reports from local media, the rejection stemmed from differing viewpoints regarding post-acquisition business strategies and organizational structure. Meta’s pursuit of AI chip manufacturers like FuriosaAI underscores its broader objective of lessening its dependence on external suppliers, such as Nvidia.

CEO June Paik on the Meta Deal

When questioned about the reasons for the failed Meta deal, FuriosaAI CEO June Paik explained to TechCrunch: “Our mission remains paramount, and this presents a compelling opportunity. Contributing to more sustainable AI computing is both personally and professionally impactful for our company.”

Paik refrained from commenting on whether the startup is currently seeking new funding, following the conclusion of potential acquisition talks.

Expanding Beyond South Korea

Paik indicated that this new partnership will unlock significant business opportunities extending beyond the borders of South Korea.

“LG AI’s EXAONE is recognized as a leading sovereign AI model within South Korea. Its usage will not be limited to LG’s internal operations. It is poised to become a primary AI model throughout the Korean AI ecosystem. We anticipate substantial demand for both EXAONE and our chip solutions, not only domestically but also internationally. LG’s team actively collaborates with and serves global clients, and we expect those customers to also utilize these technologies,” Paik stated.

A Vote of Confidence Against Nvidia

LG AI’s choice to integrate FuriosaAI’s AI chip and accelerator is particularly noteworthy as it represents one of the few public endorsements of an alternative to Nvidia by a major corporation, according to Paik. A key factor in this decision was the lower overall cost of the startup’s hardware.

“Demonstrating that our solution delivers not only robust performance but also reduces total cost of ownership was crucial,” Paik explained.

Performance and Efficiency Gains

FuriosaAI asserts that its RNGD accelerator surpassed the performance of competing GPUs when used with LG AI Research’s EXAONE models, achieving 2.25 times greater inference performance. LG also reportedly found the FuriosaAI hardware to be more energy efficient.

Purpose-Built AI Architecture

Furiosa’s chip is not a general-purpose GPU; it was specifically designed for AI workloads. “We can efficiently support a broad spectrum of AI models. However, unlike GPUs, which are fundamentally general-purpose processors, our architecture is inherently optimized for AI computing. Our chip development does not focus on rendering or mining,” Paik clarified.

Two-Year Evaluation Process

LG AI Research initiated the evaluation of FuriosaAI’s RNGD accelerator two years ago, with the goal of assessing its efficiency and determining its potential integration into EXAONE-powered services across LG’s diverse ecosystem, Paik added.

A Lean Global Team

The Seoul-based startup, which also maintains an office in Santa Clara, operates with a compact global team of just 15 employees.