Infinigence AI, a representative Chinese company in AI-native infrastructure, has completed a Series C financing round of 2.3 billion yuan (about 323.7 million US dollars), bringing its total funding since founding in 2023 to 4.3 billion yuan.
The startup, spun off from Tsinghua University and Shanghai Jiao Tong University, has attracted a wide range of backers including state-owned funds, industry giants such as Tencent, Baidu and Lenovo, and financial investors including Sequoia China.
The company says its computing aggregation platform has deployed access to 37,000 P of compute covering 16 mainstream chip types. It previously released the world's first thousand-card heterogeneous chip mixed-training platform with ten-thousand-card scalability, having achieved large-model mixed training that combines four domestic chips, from Huawei Ascend, Iluvatar CoreX, MetaX and Moore Threads, with AMD and NVIDIA chips, reaching compute utilization as high as 97.6 percent.
In cross-region scenarios, together with China Telecom it completed China's first large-scale cross-domain mixed training verification over 4,000 kilometers, running a 30-billion-parameter model efficiently between a training cluster in Hami, Xinjiang and an inference cluster in Shenzhen, Guangdong.
In July 2026 at WAIC, the company unveiled a cross-cluster heterogeneous prefill-decode separation architecture that cuts first-token latency by 51.5 percent and per-token cost by 37.5 percent. Daily token calls on its Agentic MaaS platform have grown 40-fold compared with December last year, as the company drives a tenfold reduction in inference costs.
Source: China Economic Net (ce.cn), October 12, 2026.
