Korean AI Chip Startup FuriosaAI Pursues $500M Funding Round Ahead of 2027 IPO

Strategic Funding and IPO Plans

The Seoul-based company has appointed Morgan Stanley and Mirae Asset Securities as co-advisers for the funding round, which targets between $300 million and $500 million. The capital will be allocated towards mass production of FuriosaAI's second-generation RNGD chip, development of its third-generation processor, and global business expansion. Industry estimates suggest the new funding round could push the company's valuation above 2 trillion Korean won, approximately $1.4 billion.

FuriosaAI previously raised $125 million in a Series C bridge round in July 2025, bringing its total funding to $246 million at a valuation of $735 million. The company was founded in 2017 by CEO June Paik, a veteran of Samsung Electronics and AMD.

Rejecting Meta for Independence

FuriosaAI's ambitious fundraising follows its rejection of Meta's $800 million acquisition offer in early 2025. Negotiations broke down not over price, but over disagreements about post-acquisition business strategy and organizational structure. The leadership team opted to preserve autonomy over its chip development roadmap rather than integrate into Meta's operations. When asked about the decision, CEO June Paik stated the company wanted to continue its mission and saw the independent path as an exciting opportunity.

Major Partnership Wins

Since turning down Meta's offer, FuriosaAI has secured partnerships with prominent AI players. LG AI Research adopted RNGD chips for its EXAONE large language model platform after a seven-month performance evaluation across multiple hardware options. Kijeong Jeon, product unit leader at LG AI Research, stated that after extensively testing a wide range of options, they found RNGD to be a highly effective solution for deploying EXAONE models. The collaboration targets key sectors including electronics, finance, telecommunications, and biotechnology.

In September 2025, FuriosaAI received notable validation when it became the only hardware company invited to demonstrate at OpenAI's Seoul office opening. At the event, FuriosaAI's chips powered a real-time chatbot running the 120-billion-parameter GPT-OSS model, showcasing the technology's capabilities to one of the most influential organizations in artificial intelligence.

Technical Innovation and Architecture

The RNGD chip, manufactured using TSMC's 5-nanometer process, employs a proprietary Tensor Contraction Processor architecture. Unlike traditional AI accelerators that use matrix multiplication as their fundamental operation, RNGD is built around tensor contraction operations as its primitive computing element. This architectural choice enables significantly improved efficiency for AI inference workloads.

The chip delivers 512 teraflops of FP8 performance with 512 trillion operations per second at INT8 precision and 1,024 trillion operations at INT4 precision. FuriosaAI claims the architecture delivers 2.25 times better inference performance per watt compared to traditional GPUs. The chip's thermal design power is just 180 watts, allowing deployment in standard air-cooled data centers without expensive infrastructure upgrades or specialized cooling systems.

Hardware Specifications

RNGD features eight processing elements, each containing 64 slices that can function independently or be fused in groups of up to four to act as larger processing elements. The chip is configured with two HBM3 memory modules providing 48GB of total capacity and 1.5 terabytes per second of memory bandwidth. It supports PCIe Gen5 x16 connectivity and uses 2.5D packaging with HBM modules and the system-on-chip integrated on a single interposer using the newest 12-layer HBM3 technology.

For multi-tenant environments such as Kubernetes deployments, a single RNGD chip can be partitioned to work as 2, 4, or 8 individual processing units, each fully isolated with its own cores and dedicated memory bandwidth.

Market Context and Competition

The fundraising effort comes amid heightened investor interest in AI chip alternatives, driven by surging demand for artificial intelligence infrastructure and compute capacity. The competitive landscape includes companies like Groq, which secured $750 million at a $6.9 billion valuation, and Cerebras Systems, which is pursuing $1 billion in new funding.

FuriosaAI's approach focuses on energy efficiency and ease of deployment rather than raw performance maximums, addressing practical concerns around power consumption, cooling requirements, and total cost of ownership in large-scale AI deployments. This positioning could appeal to enterprises and cloud providers looking to balance performance with operational efficiency as they scale AI workloads.

The company's planned 2027 IPO would mark a significant milestone for the South Korean AI hardware sector and provide another public market option for investors seeking exposure to the AI infrastructure buildout beyond the dominant market leader.