Korean startup FuriosaAI, dedicated to the manufacturing of processors for Artificial Intelligence (AI) applications, has rejected an acquisition offer valued at US$800 million by Meta, the parent company of Facebook, Instagram and WhatsApp.

South Korean media reported in mid-February that Meta planned to acquire the chip development company FuriosaAI to obtain its technology for AI accelerators, such as the Renegade (RNGD) chip, and drive its goals of developing customized components.

It was initially thought that the acquisition could be finalized by Meta during February or March, although more organizations were interested in acquiring the company, founded in 2017 by engineer June Paik, who previously worked at AMD and Samsung.

Unacceptable business scenarios

Now, South Korean media reports that merger and acquisition negotiations between FuriosaAI and Meta have failed. Instead of selling its management rights, the startup has opted to follow its own path and establish itself as an alternative to Nvidia in the global AI semiconductor market.

There was some internal scrutiny within FuriosaAI regarding the proposed price, but it was reported that CEO Baek-jun-ho, the founder, did not accept the conceived business scenario, which would have led to a US$800 million acquisition offer.

South Korea’s push to build a competitive domestic chip industry has significantly strengthened the presence of companies like FuriosaAI and Rebellion. Backed by a government initiative investing over US$800 million in R&D over the next five years, the country aims to boost the market share of Korean AI chips in domestic data centres from nearly zero to 80% by 2030, positioning itself as a competitive player in the global semiconductor race.

FuriosaAI’s Renegade chip tops cost-efficiency charts

FuriosaAI is currently conducting performance tests on its RNGD chip, which consumes only 25% of the energy compared to existing GPUs (having a 181W power profile per card, according to the company) in collaboration with partners such as LG, through its AI Research Center department, and Saudi Aramco.

Moreover, RNGD is built on an AI-native Tensor Contraction Processor architecture, delivering up to 3,300 tokens per second on Llama 3.1-8B models and sustaining 40 to 60 tokens per second in single-user setups. The company explains that the chip is also enhanced by an SDK featuring tensor parallelism and advanced optimization techniques, enabling scalable, multi-user deployments for complex large language and multimodal models.

Likewise, FuriosaAI plans to begin large-scale mass production of this component in the second half of the year and establish a sales base in the AI semiconductor market. The company states on its website that its mission is to “build AI chips capable of running the world’s most advanced models efficiently.”

“With RNGD now in customers’ hands, we are accelerating the next generation of frontier LLMs to unlock emerging Agentic AI applications, bringing advanced reasoning capabilities to enterprise verticals, all at dramatically lower costs,” recently said June Paik, co-founder and CEO of FuriosaAI.