Infinigence AI raises RMB 2.3B Series C: the mixed-chip compute bet keeps paying
Chinese AI infrastructure startup Infinigence AI has closed a RMB 2.3 billion (about $320 million) Series C, the company said on October 12, bringing its total financing to RMB 4.3 billion — and a new bet that inference economics, not models, is the next bottleneck.
Chinese AI infrastructure provider Infinigence AI has completed a RMB 2.3 billion (about $320 million) Series C funding round, according to company materials provided on October 12, bringing its cumulative financing to RMB 4.3 billion. The raise is another sign that investors increasingly see the next AI bottleneck in compute infrastructure, not in models themselves.
Founded in 2023, Infinigence develops software and infrastructure that let AI models run efficiently across different chip platforms. Its platform connects 37,000 petaflops of computing capacity across 16 mainstream chip types and supports mixed-chip training using both Chinese and international processors, TechNode reports.
The mixed-chip bet
In a market where chip supply is fragmented — Nvidia GPUs on one side, a wave of domestic Chinese accelerators on the other — the scarce resource is the software layer that makes heterogeneous clusters behave like a single machine. Infinigence also provides AI model inference services and infrastructure solutions for enterprise AI agents, with applications in gaming, healthcare, legal services, and energy.
The numbers behind the raise
Infinigence says its model-service platform’s daily token usage grew 40-fold by July compared with December 2025. Its cross-cluster inference architecture cut the time to generate the first token by 51.5% and the cost per token by 37.5%, according to company figures — inference economics, not just raw flops, are the product here.
The company plans to use the new capital to scale its model-service platform and further optimize its cross-cluster inference architecture, according to The SaaS News, which cites the TechNode report. The specific investors in this round were not disclosed.
Why it matters
The raise fits a 2026 pattern: money is moving down the stack. With frontier weights increasingly commoditized and open-weight models closing the gap, investors are paying for the plumbing — orchestration, inference economics, and data. Pooling 37,000 petaflops across 16 chip types is also a hedge against supply-chain fragmentation: train wherever the silicon happens to be.
The usual caveats apply — the performance figures come from the company, the investor roster is undisclosed, and China-focused compute carries geopolitical overhang. But RMB 2.3 billion is real capital, and it says inference economics is now a fundraise-tier problem.