智东西（公众号：zhidxcom）
作者 | 李水青
编辑 | 心缘
智东西10月26日消息，近日，北京AI Token生产服务商趋境科技（Approaching.AI）完成B轮融资，本轮融资金额超13亿元人民币；今年内已累计募得24亿元人民币。
本轮融资由浙财创新、萧山资本、工银投资、中关村科学城、星连资本及钱江世纪城共同领投。云创资本、湖南高新纵横、建投华科、申万创新投、中信聚信、珠海科技产业集团、商汤科技、网宿科技、甘肃树帜、常金控、元智未来、清证星竹、小咖资本等跟投，老股东尚势资本继续追投。
这是继今年7月完成A轮融资后，趋境科技短期内再获大额融资。据悉，过去半年，趋境科技多项核心经营指标在半年内实现超十倍速增长。
本轮募集的资金将重点投向高品质AI Token产能扩张、ATaaS（Approaching.AI Token as a Service）平台升级迭代、推理底层技术和新产品的研发，以及商业化合作与交付体系完善等方面。趋境科技将“高品质AI Token”定义为连接算力、模型与应用结果的最小生产单元。
趋境科技成立于2024年，坐落于北京，创始团队核心成员来自清华大学计算机系高性能所。其中，中国工程院院士郑纬民担任首席科学顾问，清华大学武永卫教授出任首席科学家，清华大学计算机系副教授章明星为公司联合发起人。
趋境科技CEO艾智远拥有清华大学计算机系博士学位，其博士阶段的研究方向主要集中在并行计算、分布式存储、AI计算引擎优化等领域。其从业经历丰富，曾担任深信服大数据、数字化、AI应用等多个部门的研发主管。
▲趋境科技CEO艾智远
在技术路线方面，趋境提出“少模型、深优化、高品质”：“少模型、深优化”解决的是“同一批算力能够生产多少高品质 AI Token”，稳定生产负载和跨集群运营解决的是“这些算力有多少时间真正处于有效工作状态”。“少模型”本身并不天然带来更高的利用率，只有单位算力产出和有效利用率同时提升，AI Token业务的单位经济才有可能成立。
目前，趋境科技已经形成一套可复制的高品质AI Token工厂建设与运营模式。半年来，伴随多项首创技术持续推出与工程打磨，公司平均单台算力的AI Token生产效率逐月攀升，高品质AI Token日均产量增长超过百倍，并攀升至超万亿量级，年内业务收入已达到去年全年近十倍。整体业务已经跨过成本线。
面对国内外芯片的性能差异，趋境科技提出了“国产Prefill-Decode（PD）异构协同”“高性能异构KV Cache转换”“异构算力计算融合”等一系列创新技术方案，重构国产算力计算模式和推理体系，并已在高标准生产场景中正式投产。
在真实生产级环境中，其实现国产芯片承载万亿级头部大模型，并达成高品质AI Token规模化生产，同时整体性价比超越国际最领先算力。
2026年，国产Token工厂成为最吸金的AI Infra赛道之一，清华系创业者成为一大主力军。
此前10月12日，上海AI Infra企业无问芯穹确认已完成23亿元C轮融资，累计融资额达43亿元，单轮融资额与累计融资额均创国内AI Infra行业新高，并正在筹备IPO；同日，同是清华系的北京AI Infra创企清程极智完成A轮融资，由中国电信和瀚晖资本联合领投。此前9月20日，已向港交所递交上市申请的北京AI Infra创企硅基流动，宣布已先后完成B+轮二期和C轮融资，2026年度累计融资金额近29亿元。
本文由 AI宝库自动抓取并翻译，版权归原作者所有；如需查看原文请点下方链接。
Intellectual (WeChat Official Account:zhidxcom)
The author, Li Shui-young
Edition.
According to the news of October 26th, recently, Beijing ' s AI Token production service provider Approaching.AI has completed the B round of financing, which amounts to over 1.3 billion yuan yuan; a cumulative 2.4 billion yuan has been raised this year.
The current round of financing was co-financed by Zhefu Innovation, Xiaoshan Capital, the Bank of China and China Bank Investment, the Science City of Central Guanamura, the Star Company Capital and Chan Gang Century City. The old shareholders continue to invest in capital.
Following the completion of round A in July of this year, it is in the short term that a significant amount of financing will be available for convergence technologies. It is known that, over the past six months, several core S & T performance indicators have grown at a rate of more than tenfold in the past six months.
This round will focus on the expansion of high-quality AI Token production, the upgrading of the Ataas (Approaching.AI Token as a Service) platform, the development of underlying reasoning techniques and new products, and the improvement of commercialization cooperation and delivery systems. Convergence technology defines "high-quality AI Token" as the smallest production unit that connects calculus, models and application results.
Convergence Technology was established in 2024 and sits in Beijing, with the core members of the founding team coming from the High Performance Institute of the University of Tsinghua Computers. Among them, Zheng Liangmin, a member of the Chinese Engineering College, served as the chief science adviser, Professor Wu Yong Wei of Qinghua University, became the chief scientist, and Zhang Star, an associate professor at the Department of Computers of Qinghua University, was co-sponsor of the company.
The Convergence Technology CEO-I-Ai has a Ph.D. in the Department of Computers of the University of Qinghua, and the course of research at the doctoral level is concentrated in the fields of parallel computing, distributed storage and optimization of the AI computing engine. It has extensive experience and has served as a research and development manager in a number of sectors, including those convinced of big data, digitization and AI applications.
It's CEO.
In terms of technology routes, the convergence suggests that “a few models, deep optimization, high quality”: “a few models, deep optimization” solves “how many high-quality AI Tokens can be produced by the same calculus”, and that “how much time do these algorithms actually work effectively” by stabilizing production loads and cross-cluster operations. The “smaller models” themselves did not naturally lead to higher utilization rates, but the unit economy of AI Token's operations could only be established if unit output and efficiency utilization were increased at the same time.
At present, convergence technology has developed a replicable high-quality AI Token plant construction and operation model. Over the past six months, with the continued roll-out of several of the first-in technologies and the grinding of works, the company ' s average single-desk-based AI Token production efficiency has risen every month, high-quality AI Token ' s daily production has increased by more than a hundred times and has risen to over trillions of quantities, with annual operating income nearly 10 times the year as a whole. The overall business has crossed the cost line.
In the face of differences in the performance of domestic and foreign chips, convergence technology has proposed a series of innovative technology options, such as the “Prefil-Decode (PD) isomeral synergy of national production” and the “High Performance Isomer KV Cache conversion” “Integration of Isomer Calculus”, to restructure the system of computational calculations and reasoning of national production, and has been officially launched in high-standard production scenarios.
In a real-production-level environment, it achieves a trillion-class head LLM on national chips and a high-quality AI Token-scale production, while the overall value for money exceeds international lead.
In 2026, the national Token factory became one of the most gold-smoking AI Infra tracks, and Qinghua entrepreneurs became a major force.
Prior to this, on 12 October, the Shanghai AI Infra firm unchartedly confirmed the completion of the 2.3 billion yuan C round of financing, the cumulative amount of which amounted to 4.3 billion yuan, the single-wheel financing and the cumulative amount of financing had reached a new high in the domestic AI Infra industry, and preparations were under way for the IPO; on the same day, the Beijing-based AI Infra venture of Qinghua was so resourceful that it completed the A round of financing, which was co-financed by China's telecommunications and capital. On 20 September, the Beijing AI Infra, which had submitted a listing application to Hong Kong Treasury, created a Silicon-based flow, announcing the completion of the B+ and C rounds of financing, with a cumulative amount of nearly $2.9 billion in 2026.
Auto-fetched and translated by AI Gem Box. All rights belong to the original author.