Xinglian Capital's portfolio company Qujing Technology has completed a Pre-A round of financing of several hundred million yuan to accelerate the construction of high-quality AI Token production infrastructure.
Xinglian Capital is a venture capital fund focusing on the large model ecosystem, with an emphasis on early - stage investments and industry connections.
Recently, Approaching.AI, an investee of Xinglian Capital, announced the completion of a Pre - A round of financing worth hundreds of millions of yuan. This round of financing was jointly led by Xinglian Capital and Huakong Technology, with follow - on investments from institutions such as Honghui Capital, Tianhao Energy, Shangshi Capital, Tianjin Ren'ai Hongsheng, and Hangzhou Fucheng. The existing shareholder, GL Ventures, continued to increase its investment.
After the completion of this round of financing, Approaching.AI will continue to increase its investment in the high - performance artificial intelligence Token production service platform (Approaching AI Token as a Service, ATaaS). It will focus on promoting the construction of computing power reserves and the underlying inference system, and continuously deliver model output capabilities with low latency, high throughput, stable structured output, reliable function calls, and predictable service quality. It will further enhance the large - scale supply capacity of high - quality Tokens for the enterprise production environment.
Approaching.AI: The core domestic supplier of high - quality Tokens, with nearly one trillion daily Token calls on the ATaaS platform
As large model applications enter the enterprise production environment, the evaluation criteria for AI inference infrastructure are changing. Enterprises no longer only focus on computing power scale, the number of models, and interface richness, but rather place more emphasis on whether each call can complete business delivery stably, efficiently, and predictably.
At this stage, the core competitiveness of inference services is shifting from "providing models" to "producing high - quality Tokens". The first Token return latency, the number of Tokens output per second, the stability of structured output, the reliability of function calls, and the predictability of service quality in high - concurrency scenarios are becoming important indicators for enterprises to choose AI infrastructure.
Approaching.AI believes that Tokens are not just the basic units of input and output for large models, but are key production factors connecting model capabilities, system performance, service stability, and cost - efficiency. Based on this judgment, the company proposed the industrial concept of Token as a Service (TaaS) and built the high - performance artificial intelligence Token production service platform ATaaS. Compared with traditional MaaS, which focuses on model calls and management, ATaaS focuses more on the delivery of inference efficiency in enterprise - level production scenarios, helping enterprises obtain the ability to produce high - quality Tokens on a large scale and in an operable manner.
In terms of model strategy, Approaching.AI adheres to the route of "fewer models, in - depth optimization". It does not aim to support hundreds of models in a general way, but focuses on a few high - productivity models and continuously optimizes the output quality, inference efficiency, TTFT stability, and TPS performance around real enterprise scenarios. For enterprise customers, the number of models does not directly equal productivity. The real key is whether each call can stably support business results.
In terms of system capabilities, Approaching.AI converts the underlying computing power into sustainable high - quality AI Token production capacity through capabilities such as heterogeneous computing power scheduling, cross - cluster cache sharing, inference link isolation, elastic scaling, and quality monitoring. Relying on its full - link system engineering capabilities, the company can provide enterprises with more stable TTFT, high - speed output capabilities of 30 - 50 TPS, and reliable service guarantees while keeping costs under control.
Currently, Approaching.AI has provided services for multiple enterprise - level customers such as Zhipu GLM and Kimi through the ATaaS platform, and the platform processes nearly one trillion Tokens per day. After long - term verification in high - complexity and high - concurrency business scenarios, the company has formed core capabilities for large - scale inference delivery.
An irreplicable talent matrix, jointly escorting the strategic development of ATaaS with business traction and a technical foundation
TaaS is not an ordinary application - layer product, but a systematic ability for the entire AI inference link. It requires the team to understand the needs of enterprise customers, industrial resources, capital paths, and commercialization rhythms, and also requires long - term accumulation in the underlying architecture fields such as computing, storage, scheduling, caching, and inference systems. The core team of Approaching.AI has both business implementation capabilities and in - depth technical R & D capabilities, laying the foundation for ATaaS to move from a cutting - edge concept to large - scale deployment among top customers within two years of its establishment.
At the business and operation level, Approaching.AI has formed the organizational ability to promote the synergy between technology productization and business capitalization. AI Zhiyuan, the founder and CEO, is a Ph.D. in computer science from Tsinghua University. He has both system research capabilities and commercialization experience in large companies. He proposed the industrial logic of TaaS and promoted the transformation of ATaaS from a technical platform to an enterprise - level production service. Dr. Wu Wenjie, the president, has a Ph.D. in finance and CFA qualifications. He has the resume of senior executives in leading industrial and capital institutions. He has led the investment and mergers and acquisitions of dozens of benchmark enterprises and comprehensively coordinates the company's strategy, internal control, and global operations. Ren Xuyang, the chairman, is an early pioneer of Baidu. He has led the establishment of companies such as iQiyi, Yidian Zixun, Haizhi, and News Break, and provides support for the company's development in terms of industrial judgment, organizational construction, capital coordination, and ecological resource integration.
At the technical and scientific research level, Approaching.AI relies on the more than 20 - year technical accumulation of the High - Performance Computing Institute of Tsinghua University and has completed the process of capital increase and shareholding with relevant technical achievements of Tsinghua University. These technical achievements were developed by scientific research teams led by Academician Zheng Weimin, Professor Wu Yongwei, and Associate Professor Zhang Mingxing over a long period, covering key directions such as high - performance computing, parallel and distributed systems, storage systems, intelligent computing power systems, and large - model inference infrastructure. The injection of these relevant achievements marks that the company's industry - university - research collaboration with the Tsinghua scientific research team in the field of AI infrastructure has entered the stage of actual implementation.
Among them, Academician Zheng Weimin, the chief scientific advisor of Approaching.AI, laid the academic foundation for high - performance computing at Tsinghua University; Professor Wu Yongwei, the chief scientist of Approaching.AI, has long been engaged in distributed and storage systems and has won many national - level science and technology awards; Associate Professor Zhang Mingxing focuses on large - model inference architectures, and the open - source projects he led, such as KTransformers and Mooncake, have been widely used in the industry. Relying on the investment of core Tsinghua technical achievements and the continuous support of top scientific research teams, Approaching.AI has formed a barrier for system engineering and scientific research transformation in the field of AI inference infrastructure.
This technical accumulation has also been verified in the open - source ecosystem. KTransformers, jointly open - sourced by Approaching.AI and the Tsinghua team, is the world's first edge heterogeneous inference framework, with more than 17k GitHub Stars, and has become the first - recommended inference engine for top large models such as GLM, Kimi, Minimax, and Qwen. In the field of distributed inference, Approaching.AI, together with Tsinghua University, Kimi, 9#AISoft, Alibaba Cloud, Ant Group and other industry - university - research institutions, co - built the open - source project Mooncake; members such as Yang Ke, a technical expert of Approaching.AI and a Ph.D. from Tsinghua University, as core contributors, are deeply involved in the implementation of multiple key technologies and architecture construction. In addition, Approaching.AI also actively participates in the technical contributions of global inference communities such as SGLang, vLLM, and NVIDIA Dynamo, continuously promoting the development of an open ecosystem for AI inference infrastructure.
The business team's understanding of industrial needs, customer scenarios, and capital paths, combined with the long - term accumulation of the technical team in the fields of high - performance computing, distributed systems, and large - model inference infrastructure, enables Approaching.AI to have the complete ability from underlying system R & D to large - scale enterprise - level delivery. As ATaaS continues to evolve, this composite team structure will continue to support the company in improving its large - scale production and delivery capabilities of high - quality Tokens.
Investors' comments
Zhang Yang, the chairman of Huakong Fund said:
With the rapid improvement of the capabilities of domestic large models and the full - scale explosion of application demand, the massive demand for Tokens is reshaping the computing power industry chain. Huakong Fund firmly believes that the AI Infra industry, which can stably provide high - quality Tokens on a large scale, will become the key infrastructure for the booming development of the AI industry, with broad market space and high investment value. The Approaching.AI team comes from the High - Performance Computing Institute of Tsinghua University, with profound scientific research background and solid technology. They have successfully connected the underlying computing power islands and have been highly recognized by the upstream and downstream ecosystems. Huakong Fund will rely on its rich AI industrial ecological resources to fully support Approaching.AI's subsequent financing and listing process and accompany the enterprise to grow into a globally leading new - generation intelligent computing "Token factory".
Li Wenjue, a partner of Xinglian Capital said:
Approaching.AI has demonstrated excellent technical depth and engineering capabilities in the field of AI infrastructure, especially leading the world in Token production efficiency. This round of investment values the systematic breakthrough of its ATaaS platform and the company's ability to quickly transform top - notch academic achievements into large - scale commercial implementation. As the application of AI Agents becomes more popular and the demand for Tokens surges, enterprises that can efficiently convert computing power into intelligent output will become the core of industrial competition. Xinglian Capital believes that in the future, those who can optimize the synergy between main control, scheduling, runtime, memory management, and model structure are more likely to master the right to speak in the next - generation AI infrastructure. Approaching.AI has both the top - notch technical background of Tsinghua University and rich commercialization experience, which makes investors full of confidence in its continuous growth and market leadership in the AI Infra track and firmly support it in building an industry benchmark for high - performance AI Token production.
This article is from the WeChat official account "Xinglian Capital", author: Xinglian Capital where good things happen. Republished by 36Kr with permission.