MatX Announces LLM Chip MatX One and Series B

MatX, building high-throughput chips for LLMs, led by Reiner Pope, Mike Gunter, Avinash Mani, and the team, has announced MatX One and a $500M Series B led by Jane Street and Situational Awareness LP. Participants include Spark Capital, Daniel Gross and Nat Friedman’s fund, Patrick and John Collison, Triatomic Capital, Harpoon Ventures, Andrej Karpathy, Dwarkesh Patel, and others, including investors across the supply chain, including Alchip and Marvell.

The MatX One chip is built on a splittable systolic array, combining the energy and area efficiency of large arrays with high utilization across smaller, flexible matrix shapes. It pairs the low latency of SRAM-first architectures with the long-context support of HBM, alongside a redesigned numerical approach.

The result: higher LLM throughput than any announced system while matching SRAM-first latency, delivering smarter, faster models per subscription dollar.

MatX was founded on the belief that the best chip for LLMs must be designed from first principles, grounded in a deep understanding of how large models operate and evolve. The company prioritizes peak LLM performance, even at the expense of small-model efficiency, low-volume workloads, or programming simplicity, to deliver a purpose-built architecture for frontier AI.