FuriousaAI Inc., a Seoul-based developer of artificial intelligence chips, is reportedly in talks to raise a new round of funding. According to sources cited by Bloomberg, the startup is seeking between $300 million and $500 million. To assist with closing this round, FuriousaAI has hired Morgan Stanley and Mirae Asset Securities, Korea’s largest investment bank.
This proposed investment round is described as a Series D raise. However, the report did not specify the anticipated valuation for FuriousaAI. As of July, the company was valued at $735 million, following a $125 million funding round from a group of institutional investors. Notably, a few months prior, FuriousaAI reportedly turned down an $800 million acquisition offer from Meta Platforms Inc.
—
### Innovating AI Chip Design: From Matrix Multiplications to Tensor Contractions
There are often multiple ways to approach a mathematical operation. For example, adding two and two produces the same result as multiplying two by two. A similar principle applies in matrix multiplications—key calculations that AI models use to process data efficiently.
A matrix is a mathematical structure composed of numbers organized into rows and columns. Matrix multiplication combines two matrices into one through a process involving multiplication and other operations.
Traditionally, server-grade graphics processing units (GPUs) contain numerous cores optimized specifically to perform matrix multiplications. FuriousaAI has taken a different approach with its flagship RNGD chip. Instead of designing the processor focused on matrix multiplications, the company optimized it for a mathematical operation called tensor contraction.
FuriousaAI claims that tensor contractions can produce the same results as matrix multiplications but more efficiently, translating into faster AI performance.
—
### How FuriousaAI’s Chip Architecture Accelerates Neural Networks
The company’s chip architecture speeds up neural networks in two main ways:
1. **Improved Parallelism:** This allows more calculations to be carried out simultaneously (side-by-side) rather than sequentially (one after another).
2. **Increased Data Reuse:** By reusing a piece of data, the chip reduces the number of times data must be retrieved from memory or generated from scratch. This helps avoid associated delays in computation.
The RNGD chip can perform up to 512 trillion calculations per second on data stored in the FP8 format. It is shipped in a PCIe card that has a thermal footprint of 180 watts, meaning it does not require water cooling.
Additionally, FuriousaAI offers a server called the NXT RNGD, which includes 8 RNGD accelerators and delivers up to 4 petaflops of performance.
—
### Looking Ahead: IPO Plans
The funding round currently underway could reportedly be FuriousaAI’s last before going public. The company is expected to list its shares in 2027, marking a significant milestone in its growth trajectory.
—
FuriousaAI’s innovative chip design and cutting-edge technology place it at the forefront of the AI hardware space, with promising prospects for investors and the AI community alike.
https://siliconangle.com/2026/01/19/ai-chip-developer-furiousaai-reportedly-raising-500m/