South Korean startup FuriosaAI has rejected an $800 million acquisition offer from Meta Platforms. The company, which introduced the RNGD AI accelerator for large language models (LLM) and multimodal AI last year, intends to continue its development independently, Bloomberg reports, citing an informed source.
Meta had reportedly been in acquisition talks with FuriosaAI since early 2025. Following speculation about a possible deal in February, shares of its largest investor, South Korean venture capital firm DSC Investment, saw an increase. However, after news of the refusal, DSC Investment’s stock dropped by 16% on Monday.
Advancing AI Hardware Development
FuriosaAI’s RNGD accelerator, built using TSMC’s 5nm process and equipped with HBM3 memory chips from SK Hynix, competes with NVIDIA and other AI startups like Groq, SambaNova, and Cerebras. The company employs approximately 150 people, including 15 staff members in its Silicon Valley office. At the end of last year, FuriosaAI began testing RNGD and distributing sample chips to customers such as LG AI Research and Saudi Aramco. More than a dozen potential clients are expected to evaluate the chips in the first half of 2025.
To support its growth, FuriosaAI is preparing to raise additional capital before pursuing an initial public offering (IPO). Sources indicate that an extended Series C funding round is expected to close within a month, securing the necessary funds. Since its founding in 2017, the startup has raised $115 million across four funding rounds, according to Data Center Dynamics.
Meta’s AI Ambitions and Future Outlook
Meta continues to invest heavily in AI infrastructure to compete with OpenAI and Google. In January 2025, CEO Mark Zuckerberg announced plans to allocate up to $65 billion for AI development, covering data centers and AI talent acquisition. Shortly after, he informed investors that the company intends to spend hundreds of billions of dollars on AI infrastructure in the long term.
As part of its AI expansion, Meta is also developing proprietary chips, adds NIX Solutions. In 2023, it introduced the Meta Training and Inference Accelerator (MTIA) for AI workloads, followed by a second-generation MTIA in 2024, which offers three times the performance of its predecessor. The company is currently testing its own AI training accelerator.
FuriosaAI’s decision to remain independent signals confidence in its technology and business strategy. While the company moves forward with expansion and funding, we’ll keep you updated on further developments.