MemSilicon raises $59.8M Series A, names chip veteran Lu Tao CEO
What's the deal? Chinese compute-in-memory chip startup MemSilicon has raised more than US$55M ($59.8 million at current rates) in a Series A round, and named former Graphcore president and Greater China head Lu Tao as chief executive officer and co-founder. The round was led by Boyuan CapitalDealroom has a profile for this one. Try Dealroom →, with Bosch VenturesDealroom has a profile for this one. Try Dealroom →, Chuangling CapitalDealroom has a profile for this one. Try Dealroom →, MiHoYoDealroom has a profile for this one. Try Dealroom →, Zhongguancun Science CityDealroom has a profile for this one. Try Dealroom →, and GigaDevice among the backers.
What's the endgame? MemSilicon builds chips using RRAM — resistive memory — as a storage foundation and an LPU as a compute engine, aimed at large-model inference. The new money funds two efforts: commercialising RRAM and developing a first-generation inference LPU.
Why now? The company says large-model inference is entering a critical phase of architectural innovation, where raw compute increases alone can't meet demands for low latency, long context, and high throughput. Lu argues compute-storage co-design is one of the core paths forward.
Lu brings more than 20 years in chips, with prior roles at Graphcore, Cavium, and Xizhi (Lightelligence), where he worked across AI infrastructure components. He joins a team originating from Chinese Academy of Sciences academician Liu Ming's group, with decades of materials and device work. "We won't fight GPUs head-on," Lu said, framing the company's focus on high-end inference and cost-efficiency.
What's the context? The Series A follows an angel round of nearly US$13.8M in 2025 and a Pre-A of more than US$13.8M in January 2026. In October, MemSilicon taped out its first high-density RRAM chip, "MountTai," built on a 28nm process with 1Gb capacity and 30Mb/mm² density; the chip is now in verification.
What could go wrong? MemSilicon has yet to demonstrate commercial-scale deployment. Moving from its core memory technology into the crowded AI-inference race — where rivals are also raising substantial capital — adds execution risk, and its LPU remains in development.
The signal: The round reflects investor appetite for specialised inference architectures as the AI race shifts from peak compute to task-completion efficiency. MemSilicon is betting that China's memory strengths, paired with Lu's systems experience, can turn lab research into shippable products.
Read more: sohu.com
Image credit: Homedust
