MoXin raises nearly RMB 1 billion in C+ round, five months after its last ~RMB 1 billion raise

量子位 · wechat · 2026-11-06

Sparse-compute chip startup MoXin has closed a C+ round of nearly RMB 1 billion, just five months after a similarly sized previous round, signaling strong investor confidence in its dedicated sparse-inference chip approach.

Founded in 2018, MoXin builds on dual weight-and-activation sparsification algorithms with algorithm-compiler-hardware co-design; its S30 card has topped select MLPerf Inference categories for three consecutive sessions. The company argues that as MoE-style conditional computation spreads, sparse-specialized silicon can convert model-level compute savings into real inference cost reductions that general GPUs struggle to match.

Commercially, its inference clusters are deployed across four Chinese regions; it joined Shanghai's offshore "Token factory" data-center project pairing wind-power green electricity with sparse-architecture cost cuts, and with PKU Lingang achieved up to 2.88x inference throughput and 7.64x model compression on remote-sensing models with minimal accuracy loss. A new SP200 board arrives by year-end alongside an upgraded SECA software stack.

Original post →

MoXin raises nearly RMB 1 billion in C+ round, five months after its last ~RMB 1 billion raise

More from Venture