Approaching.AI raises over RMB 1.3B Series B to scale premium AI token production

Approaching.AI (趋境科技), a Chinese company that sells what it calls high-quality AI token production, said it has closed a Series B round of more than RMB 1.3 billion. Including earlier closings this year, it has raised RMB 2.4 billion in 2026, at a pace of roughly one round every two months.

The round was led by Zhecai Innovation, Xiaoshan Capital, ICBC Investment, Zhongguancun Science City, Xinglian Capital and Qianjiang Century City. Yuncheng Capital, Hunan Gaoxin Zongheng, Jiantou Huake, Shenwan Innovation Investment, CITIC Juxin, Zhuhai Technology Industry Group, SenseTime and Wangsu Science and Technology joined, while existing investor Shangshi Capital increased its stake. The company had only closed its Series A in July.

The proceeds are earmarked for capacity rather than pure research. Approaching.AI says the money will go toward expanding high-quality token capacity, upgrading its ATaaS (Approaching.AI Token as a Service) platform, developing inference-level technology and new products, and building out commercial delivery.

The company positions itself around a deliberate constraint: fewer models, deeper optimisation, higher quality. Instead of supporting as many models as possible, it concentrates engineering on a small set of widely used mainstream models, arguing that enterprise customers pay for business outcomes rather than for the number of models a platform can host.

On the technical side, Approaching.AI says it has put heterogeneous inference, storage-for-compute substitution and virtual-real co-architecture into production, alongside domestically built Prefill-Decode heterogeneous coordination, high-performance heterogeneous KV cache conversion and cross-chip compute fusion. The claim is that trillion-parameter frontier models can run on chips of different domestic architectures at production-grade quality.

The operating numbers disclosed alongside the funding are what give the round its weight. The company says daily output of high-quality AI tokens rose more than a hundredfold in six months to the trillion-token level, while revenue this year is close to ten times its full-year total for last year, and that the business has crossed its cost line. Projects already in production carry daily capacity in the trillions of tokens, with more than ten trillion-token-per-day factory projects under way.

The rollout is regional as much as technical. In Henan the company operates alongside local state capital to produce, sell and absorb tokens; its east China headquarters in Hangzhou's Qianjiang Century City is planned as a ten-thousand-GPU-class token factory; in Hainan it is testing a model where local state capital builds the facility and Approaching.AI supplies the platform and operations. On the ecosystem side it is building a trillion-token-per-day plant with SenseTime, a trillion-token-per-day facility with Paratera, and working with Hongxin Electronics, Wangsu, Moore Threads, MetaX and Tianshu Zhixin on chips and delivery.

The team behind the pitch includes founder and CEO Ai Zhiyuan, president and CFO Wu Wenjie and CTO Chen Xianglin, with academician Zheng Weimin as chief scientific adviser and professor Wu Yongwei as chief scientist. The team led the creation of KTransformers and took part in founding and building Mooncake, two open-source projects that anchor its inference engineering credentials.

What matters here is that the funding story has shifted from renting out compute to the delivery quality and unit economics of token factories, with investors pricing orders, capacity and cost lines rather than parameter counts. The things to watch are whether the ten-thousand-GPU-class plants land on schedule, whether the domestic heterogeneous stack reproduces its cost advantage on more top-tier customer workloads, and whether the two-month funding cadence continues as capacity expands.