Shenzhen diffusion language model startup DiffuSpace raises two rounds totaling about $70M

DiffuSpace said on October 9 that it closed two back-to-back rounds worth several hundred million yuan. It did not give an exact figure, but investors told Chinese outlet Touzhong the total was close to RMB 500M, about $70M. Matrix Partners China, Shunwei Capital and Legend Capital co-led, joined by CAS Star, Huawei Hubble and Horizon Robotics. The money goes to model training, infrastructure and vertical adaptation. The company was founded in Shenzhen in May 2026 by Lingpeng Kong, co-director of the University of Hong Kong NLP lab, with PhD students Shansan Gong and Jiacheng Ye. The team built DiffuSeq and Dream 7B and plans to release a larger open-weight model soon. It already works with agent platform Acrab on on-device deployments. Diffusion language models generate many tokens in parallel instead of one at a time, which can cut inference latency sharply. Google, Inception and Ant Group are exploring the approach. A well-funded open-weight dLLM lab could make fast on-device generation a practical alternative for builders.