SUNMMIO completes Pre-A2 and Series A for 3D-stacked AI inference chips

Beijing-based SUNMMIO, known in China as 算苗科技, announced the two financings with backing from government-linked platforms including Yizhuang State-Owned Capital Investment. The company says it raised nearly RMB 2B across multiple rounds in the past year. That figure covers all rounds, so the amount tied to this announcement is unknown. No valuation or lead investor was disclosed. Its 3D TokenPU design stacks memory wafers above compute circuitry to shorten data paths and ease the memory limits that drive inference cost. The first processor, the A4E, taped out in June and is designed for 16 terabytes per second of memory bandwidth. Tapeout and specifications alone do not show yields, pricing or performance with real customers. For AI builders, the round shows Chinese state-linked capital funding inference-specific architectures as an alternative to conventional GPUs. Memory bandwidth is the bottleneck for serving large models, so alternative designs that work could shift inference pricing, although silicon risk is high until volume parts ship.