American startup MatX, which specializes in creating chips for large language models (LLM), has raised $500 million in a Series B funding round. The company aims to become a serious competitor to Nvidia in the field of artificial intelligence and accelerators for AI applications.
This is reported by Finway
Investors and Development Strategy
The round was led by Jane Street and the Situational Awareness investment fund, founded by former OpenAI researcher Leopold Ashenbrenner. The list of investors also includes Spark Capital, Marvell Technology, NFDG, Stripe co-founders Patrick and John Collison, as well as several prominent private technology investors.
MatX was founded in 2023 by two former Google engineers who previously worked on the TPU project—Google’s proprietary AI chips. The company’s CEO, Rainer Pope, was responsible for software development for processors, while his partner Mike Gunter handled the hardware aspect.
The startup announced the development of the MatX One chip, which, according to the company, will be ten times more efficient than current Nvidia GPUs when training LLMs. The first shipments of MatX One are scheduled for 2027, with production to be carried out in collaboration with TSMC.
“MatX promises a chip that is 10 times more efficient than Nvidia GPUs during LLM training.”
Technological Features and Market Prospects
The MatX One is being specifically designed for large language model tasks, providing a significant performance boost compared to traditional graphics processors. The main focus is on high bandwidth and minimal latency, which enhances the efficiency of LLM training and inference, reduces energy consumption, and optimally utilizes chip area.
The chip’s architecture is based on a modified systolic structure, combining elements from SRAM memory solutions and supporting long contexts through HBM memory. This approach allows the startup to optimize chips for large language models, even at the expense of versatility and support for smaller systems.
The current valuation of the company is not disclosed; however, in 2024, during a previous funding round, MatX raised approximately $100 million at a valuation of over $300 million. In comparison, competitor Etched raised $500 million in January 2026 at a valuation of around $5 billion.
Given the rapid growth in demand for AI accelerators and the shortage of Nvidia GPUs, investors are showing particular interest in alternative architectures. MatX plans to complete product development within a year and scale production. The project is being worked on by a team of about 100 engineers based in a single research center.