Demand is not what troubles Chinese AI buyers; supply is. Export rules keep redrawing which chips are available, so anchoring a training cluster to one processor family amounts to a bet on future access.
Infinigence AI sells software and infrastructure that let models run efficiently across differing chip platforms. The 2023-founded company has now closed a RMB 2.3B Series C, lifting its cumulative financing to RMB 4.3B, according to materials it provided on October 12.
Spanning 16 mainstream chip types, its platform ties together 37,000 petaflops of computing capacity and enables mixed-chip training that runs Chinese and international processors inside one job. Inference and infrastructure products for enterprise AI agent deployments round out the offering, serving gaming, healthcare, legal services and energy customers.
Mixed-chip training is no small engineering feat, because parallelism strategies and toolchains vary from vendor to vendor. Cracking it lets buyers source across suppliers without rewriting their training stack. The company says daily token usage on its model-service platform rose 40-fold by July versus December 2025, and that its cross-cluster inference architecture cut time to first token by 51.5% while lowering cost per token by 37.5%.
The round joins a cluster of big Chinese AI infrastructure raises this week, among them BlueCore Compute’s roughly RMB 2B B round and more than RMB 1B Pre-A for Mosi Intelligence. Investors are crowding into the layer below the models.
