Author | Li Shuiqing Editor | Xinyuan
Zhitongshi reported on October 26 that Approaching.AI, a Beijing-based AI token production service provider, recently completed its Series B funding round, raising over CNY1.3 billion. The company has accumulated CNY 2.4 billion in fundraising this year alone. This round was jointly led by Zhejiang Finance Innovation, Xiaoshan Capital, ICBC Investment, Zhongguancun Science City, Xinglian Capital, and Qianjiang Century City. Follow-on investors included Yunchuang Capital, Hunan High-Tech Zongheng, Jian Tou Huake, Shenwan Innovation Investment, CITIC Juxin, Zhuhai Technology Industry Group, SenseTime, Wangsu Technology, Gansu Shuzhi, Changjin Control, Yuanzhi Future, Qingzheng Xingzhu, and Xiaoka Capital. Existing shareholder Shangshi Capital continued to increase its stake. This marks another large-scale financing for Approaching.AI shortly after completing its Series A round in July of this year. It is reported that over the past six months, several core operational metrics at Approaching.AI achieved more than tenfold growth.
The funds raised in this round will be primarily invested in expanding high-quality AI token capacity, upgrading and iterating the ATaaS (Approaching.AI Token as a Service) platform, R&D of underlying inference technologies and new products, as well as refining commercial cooperation and delivery systems. Approaching.AI defines "high-quality AI tokens" as the smallest production unit connecting compute power, models, and application outcomes.
Founded in 2024 and headquartered in Beijing, Approaching.AI's core founding team members come from the Institute of High-Performance Computing at Tsinghua University's Department of Computer Science. Academician Zheng Weimin serves as Chief Scientific Advisor, Professor Wu Yongwei from Tsinghua University serves as Chief Scientist, and Associate Professor Zhang Mingxing from Tsinghua University's Department of Computer Science is a co-founder. CEO Ai Zhiyuan holds a PhD from Tsinghua University's Department of Computer Science, with doctoral research focused on parallel computing, distributed storage, and AI computing engine optimization. He has extensive professional experience, having served as R&D head for multiple departments including big data, digitalization, and AI applications at Sangfor Technologies.
In terms of technical roadmap, Approaching.AI proposes "Fewer Models, Deep Optimization, High Quality": "Fewer Models, Deep Optimization" addresses how many high-quality AI tokens can be produced with a given batch of compute resources, while stable production loads and cross-cluster operations address how much time these compute resources are truly in an effective working state.
"Fewer Models" itself does not naturally lead to higher utilization rates; only when both per-unit compute output and effective utilization improve simultaneously can the unit economics of the AI token business become viable. Currently, Approaching.AI has established a replicable model for building and operating high-quality AI token factories.
Over the past six months, accompanied by the continuous release and engineering refinement of several pioneering technologies, the company's average AI token production efficiency per unit of compute has risen month by month. The daily production volume of high-quality AI tokens has grown by more than a hundredfold, reaching the trillion-level scale. Annual revenue has reached nearly ten times last year's full-year total. The overall business has crossed the cost line. Facing performance differences between domestic and international chips, Approaching.AI proposed innovative technical solutions such as "Domestic Prefill-Decode (PD) Heterogeneous Collaboration," "High-Performance Heterogeneous KV Cache Conversion," and "Heterogeneous Compute Fusion," reconstructing domestic compute modes and inference systems. These have been officially deployed in high-standard production scenarios. In real production-grade environments, the company has enabled domestic chips to support trillion-parameter leading large models, achieving scalable production of high-quality AI tokens, while surpassing the overall cost-performance ratio of the most advanced international compute resources.
In 2026, domestic token factories have become one of the most lucrative AI infrastructure tracks, with Tsinghua-affiliated entrepreneurs forming a major force. Previously, on October 12, Shanghai-based AI infrastructure company Infinigence AI confirmed it had completed a C-round financing of CNY 2.3 billion, bringing its cumulative fundraising to CNY 4.3 billion. Both the single-round and cumulative amounts set new highs in the domestic AI infrastructure industry, and the company is preparing for an IPO. On the same day, Beijing-based AI infrastructure startup Qingcheng Jizhi, also founded by Tsinghua alumni, completed its Series A round, jointly led by China Telecom and Hanhui Capital. Prior to this, on September 20, SiliconFlow, a Beijing-based AI infrastructure startup that has already submitted its listing application to the HKEX, announced it had successively completed the second phase of its B+ round and its C round, with cumulative fundraising for 2026 approaching CNY 2.9 billion.