The AI Token factory sector is booming, with Tsinghua faculty and students securing 1 billion yuan in financing within half a year

Who is producing the equivalent of "water, electricity and coal" for the AI era?

A financing deal has come into our view — on July 13, Qujing Technology's Series A financing was unveiled, with heavy lead investment from the Huirong Fund of Henan Investment Group, while existing shareholders including Zhenxin Capital, Shangshi Capital, Xinglian Capital, Shanghai Guofang Innovation, Honghui Fund, Huakong Fund, and Hangzhou Fucheng continued to make over-subscribed additional investments.

This is a team composed of Tsinghua University faculty and students: founder and CEO Ai Zhiyuan, and CTO Chen Xianglin both graduated from the High Performance Computing Institute of the Department of Computer Science and Technology of Tsinghua University; Zheng Weimin, an academician of the Chinese Academy of Engineering who also comes from the High Performance Computing Institute, serves as the chief scientific advisor; Professor Wu Yongwei of Tsinghua University acts as the chief scientist; Associate Professor Zhang Mingxing of the Department of Computer Science and Technology of Tsinghua University, as a co-initiator, has long guided the company's technical strategy and key R&D breakthroughs, continuously driving cutting-edge technological innovations.

Three years ago, most domestic AI startups focused on large language models. Even startups in the AI Infra field mostly concentrated on model training. However, Qujing Technology chose to start from the large language model inference track, building a high-quality AI Token factory. Nowadays, as the demand for AI Token grows exponentially, this once-obscure track has finally become popular. Similar to Zhipu AI, Qujing Technology has completed the technology transfer and equity participation from Tsinghua University, becoming a typical project for the transformation of Tsinghua's scientific and technological achievements.

In just half a year, Qujing Technology has accumulated financing of over 10 billion yuan. A series of 100-billion and trillion-level high-quality AI Token factories that Qujing has participated in building have been completed one after another, creating a new landscape for the AI industry.

Deep Integration of Industry and Research, Building High-Quality AI Token Factories

Back in 2023, ChatGPT ignited the global generative AI wave. Witnessing this historic opportunity, Professor Wu Yongwei from the Department of Computer Science and Technology of Tsinghua University and Ren Xuyang, founder of Zhenxin Capital, decided to jointly establish Qujing Technology. Their technical starting point is precisely the High Performance Computing Institute of Tsinghua University.

At the end of December of the same year, Qujing Technology was officially established. The company's founder and CEO, Ai Zhiyuan, holds a doctorate from the High Performance Computing Institute of Tsinghua University. He once served as the R&D head in multiple key departments of listed companies, covering big data, digitalization, and AI applications, and has accumulated complete industrial experience from technical R&D to large-scale implementation. Co-initiator Zhang Mingxing, an associate professor at Tsinghua University, mainly conducts research in the field of computer architecture and has deeply participated in the basic system construction of leading large language model manufacturers. As the company entered the stage of accelerated marketization, Dr. Wu Wenjie took office as the president of Qujing Technology in March this year. As a senior industry expert in finance and strategy with a doctorate in finance from the University of Hong Kong, she further strengthened the company's capabilities in financial management and global operations. Thus, a core team with both technical background, commercial perspective, and industrial experience was formed.

Focusing on AI, the team made a choice that was not mainstream at the time — when most AI entrepreneurs chose to invest in large language model training, Qujing Technology focused on AI inference from the very beginning. In simple terms, training is about creating a "smart brain", while inference is about how to use the brain efficiently.

"Training is a cost item, and inference is a profit-generating item," Ai Zhiyuan explained. Their judgment at that time was that inference can truly generate economic benefits and will be a broader market. What Qujing Technology aims to do is to become the best partner for the construction and operation of Token factories in the AI era, making the process of using the "brain" more efficient.

This is also the positioning of Qujing Technology — compared with other AI Token factories, Qujing Technology targets high-quality AI Token production. Ai Zhiyuan further explained that when large language models truly enter the production stage, what customers need is no longer a large model that "can chat", but one that can complete real business stably, efficiently, and at low cost.

AI Tokens that truly have enterprise-level implementation value need to continuously meet multiple requirements on models with hundreds of billions or even trillions of parameters, including low first-Token latency, high concurrency support, stable output quality, structured result generation, and function calls, while controlling the unit generation cost within an acceptable range for enterprises.

None of these capabilities is the hardest to achieve individually, which is the choice of most AI infra companies. But the real challenge is that the actual needs of customers require all these indicators to be met simultaneously under real production loads and remain stable during long-term operation. According to data calculations, different combinations of capabilities can lead to production efficiency gaps of several times or even dozens of times.

To achieve this goal, Qujing Technology has built end-to-end capabilities covering heterogeneous integration, intelligent orchestration, and elastic scaling through globally pioneering technologies such as "full-system heterogeneous collaboration", "computing via memory", and "virtual-real isomorphism". Instead of focusing on a single pain point, it optimizes every link of AI Token production, ultimately achieving an order-of-magnitude improvement in efficiency.

Based on this, Qujing Technology also proposed the Token as a Service (TaaS) concept. With the self-developed high-efficiency AI Token production service platform ATaaS as the core, it breaks the conversion bottleneck between computing hardware investment and AI Token production capacity through system architecture and engineering capabilities, continuously and stably outputting high-quality AI Tokens like a standardized production line.

In the past two years, this team has rarely appeared in the public eye. But Qujing Technology's choice is being validated by the industry — when Token becomes the currency of the AI era, the best time has finally arrived.

Explosive Business, Over 10 Billion Yuan in Financing in Half a Year

Investors began to flock to Qujing.

Looking back, in February this year, Qujing Technology completed the Angel++ round of financing invested by Parallel Technology. In May, it completed the Pre-A round of financing, further expanding the investor lineup. This round was jointly led by Xinglian Capital and Huakong Fund, with participation from institutions such as Honghui Fund, Tianhao Energy, Shangshi Capital, Tianjin Ren'ai Hongsheng, and Hangzhou Fucheng, while existing shareholder GL Ventures continued to increase its investment.

In the latest round, investor enthusiasm remains undiminished: led by the Huirong Fund of Henan Investment Group, existing shareholders including Zhenxin Capital, Shangshi Capital, Xinglian Capital, Shanghai Guofang Innovation, Honghui Fund, Huakong Fund, and Hangzhou Fucheng continued to make over-subscribed additional investments. So far, Qujing Technology has accumulated financing of over 10 billion yuan in half a year. It is learned from Pedaily that the company's next round of financing is already in progress.

As far as the eye can see, more and more mainstream institutions are choosing to bet on Qujing, placing their votes for the future in advance. The continuous additional investment from existing shareholders is the strongest endorsement of Qujing's industrial judgment, technical strength, and phased achievements.

Behind this, Qujing's initial judgment has finally come true: with the rapid popularization of AI Coding, OpenClaw and other technologies, the demand for large-scale inference is accelerating, and the commercial implementation of AI is fully exploding. The team's global leading technological innovations in the directions of computing via memory, full-system heterogeneous collaboration, and virtual-real isomorphism in the past are ushering in a window of value realization, which is ultimately reflected in the production efficiency of high-quality AI Tokens — the stronger the technical capability, the higher the inference efficiency, the lower the unit Token cost, and the greater the profit space for enterprises.

As a result, Qujing Technology is highly sought after by the venture capital circle. Investors increasingly recognize the route that the team has adhered to since its establishment — "fewer models, deeper optimization". Qujing's focus is not on expanding the number of models, but on targeting real production scenarios, selecting key large language models for in-depth refinement, and continuously improving their performance, stability, cost efficiency, orchestration capabilities, and cluster operation levels.

Ai Zhiyuan gave a vivid metaphor: compared to building a "hypermarket" that sells everything, Qujing prefers to become a "boutique specialty store". Instead of continuously expanding the number of models, the company wants to concentrate resources on a small number of high-productivity models and high-value scenarios, so that the same amount of computing power can continuously output more high-quality AI Tokens.

The underlying business judgment is: enterprise-level customers ultimately pay for business results, not for the number of compatible models. In fact, the competition pattern of large language models has gradually converged now. "Currently, less than 10% of the leading models in China occupy the vast majority of the AI Token market." Based on this judgment, Qujing concentrates its resources on a small number of leading models and core scenarios, thus realizing the compound interest effect of continuous optimization.

Pedaily obtained a set of data: since the Spring Festival in 2026, the average AI Token production efficiency of a single computing device of Qujing Technology has increased by more than 3 times, and the total output of high-quality AI Tokens has increased by more than 30 times. Among them, a leading large language model with trillion-level parameters has achieved a daily production capacity of trillion-level high-quality AI Tokens. At the same time, the revenue in the single month of June 2026 has exceeded the total revenue of 2025, and the revenue scale is still growing at a high speed.

In the view of Qujing Technology, the ultimate competition of AI infrastructure is not just about who has more GPUs, nor about how many types of models are supported, but about who can continuously produce more, more stable, and higher-quality AI Tokens. These are precisely the capabilities that investors value together.

Token is King, Embracing the New AI World

AI inference is exploding.

Data from the National Data Administration shows that as of March 2026, the average daily Token call volume in China has exceeded 140 trillion, an increase of more than a thousand times compared to two years ago. The core factor driving the explosion of AI Tokens is the overall expansion of inference demand. This also confirms the judgment of investors that AI inference will become one of the largest markets in the world.

When Token becomes the "water, electricity and coal" of the AI era, a hidden and high-growth underlying business is emerging — the "AI Token factory". The underlying logic is simple: in the future, whoever can supply high-quality AI Tokens in a lower-cost, more stable, and more controllable way will seize the opportunity in the new Token economy track.

As a result, a "race for positions" in AI Token factories is unfolding across the country. As the track becomes more and more crowded, commercialization has become a question that AI infra enterprises must answer.

To a certain extent, Qujing has greater ambitions: not just to become an AI Token factory, but the designer, builder, producer, and operator of AI Token factories, and an indispensable part of the AI Token ecosystem.

This is also reflected in Qujing's two business models: one is the direct sales model. After leasing or obtaining computing resources, Qujing directly produces high-quality AI Tokens and supplies them to leading model manufacturers, internet platforms, AI application companies, and large enterprise customers, obtaining higher returns by improving AI Token production capacity and operation efficiency. The other is the co-operation model. For customers who plan to or already have computing resources, Qujing undertakes the overall planning, system integration, construction and delivery, and subsequent co-operation of AI Token factories.

Ai Zhiyuan further explained that more and more listed companies, state-owned enterprises, and smart computing centers around the country hope to transform from traditional computing power leasing to higher value-added AI Token production. However, there are not many teams that truly have the capabilities of inference system design, heterogeneous computing, and operation. What Qujing provides is a complete set of AI Token factory design and construction solutions, helping partners complete the transformation from "selling computing power" to "selling high-quality AI Tokens".

A real AI Token economy should not be that one company completes all links alone, but that more industrial partners participate in it to jointly build the ecosystem. Obviously, Qujing has become a key part of the ecosystem, connecting industrial partners such as models, computing power, and applications, and promoting the AI industry to move from single-point innovation to ecological symbiosis.

This is also the inevitable direction of industrial evolution. Every time a technological wave truly changes the world, it often does not happen when new technologies are born, but undergoes qualitative changes because a set of infrastructure supporting its operation matures. Just as the steam engine era needed railways; the internet era needed optical fibers and data centers. The AI era also requires a new set of infrastructure to allow intelligence to flow stably, efficiently, and at low cost to all industries.

Behind this, huge business opportunities are hidden in the new AI infrastructure track. Only by deeply cultivating underlying innovation and collaborating with the industrial ecosystem can we go through the technological cycle and continuously release long-term industrial and commercial value.

This article is from the WeChat Official Account "Pedaily" (ID: pedaily2012), author: Wu Qiong, published with authorization from 36Kr.