Since mid-to-late October, the AI Infra sector has seen a funding deal land almost every two or three days, which stands out in a capital market that has broadly cooled in the second half of the year. One investor complained of being "not quite sure what to invest in next year."

Qujing Technology is the latest deal in this sector. The high-quality AI Token production services provider says it has closed a Series B of more than 1.3 billion yuan, bringing its total raised in 2026 to 2.4 billion yuan. Run through its funding history: Pre-A on May 20, Series A on July 13, Series B in October. That is a round every two months on average, with each round larger than the one before.

For Token producers, 2026 is a year of contradictions. It has been labeled both the "first year of commercialization" and the "first year of the inference application boom," even as surging demand collides with profit pressure in a shakeout. Founders who raise money still have to answer the same questions: can you deliver, what are your gross margins, and where are the orders.

Six Lead Investors, and a Name List That Doubles as a Capacity Map

The round was co-led by six institutions: Zhejiang Finance Capital, Xiaoshan Capital, ICBC Investment, Zhongguancun Science City, Xinglian Capital and Qianjiang Century City. Participating investors included Yunchuang Capital, Hunan Gaoxin Zongheng, Jiantou Huake, Shenwan Innovation Investment, CITIC Juxin, Zhuhai Technology Industry Group, SenseTime, Wangsu Science & Technology, Gansu Shuzhi, Changzhou Financial Holdings, Yuanzhi Weilai, Qingzheng Xingzhu and Xiaoka Capital, with existing shareholder Shangshi Capital adding to its stake.

The list shows local state capital, financial capital, industrial companies and market-driven funds all turning up at once.

Map the names onto geography and the picture sharpens. Zhejiang Finance Capital is a provincial financial platform in Zhejiang; ICBC Investment is the financial asset investment arm of Industrial and Commercial Bank of China; Zhongguancun Science City represents Beijing's Haidian district; Xiaoshan Capital and Qianjiang Century City are both based in Xiaoshan, Hangzhou. Scattered across the rest of the list are local capital vehicles from Hunan, Zhuhai and Gansu, plus financial capital from the Shenwan, CITIC and China Construction Bank systems. This shareholder register reads more like a capacity map.

Why the Shareholder Seats Changed Hands

The playbook already worked once at the Series A. That round, on July 13 this year, was led by Huirong Fund under Henan Investment Group, at a time when Henan and Qujing were jointly developing an AI Token factory with daily capacity in the trillions. Investors were not just buying equity; they were co-building capacity.

By the Series B, that logic is even clearer. Zhu Lifeng, chairman of Xiaoshan Capital Group, called Qujing Technology a "benchmark project for Xiaoshan's AI Infra buildout," with the two sides working together on a 10,000-chip-class AI Token factory and moving to close an industrial loop of "chip-model synergy." The industrial investors are worth a look too: SenseTime and Wangsu Science & Technology both appear among the participating investors, and both are also ecosystem partners. Qujing is working with SenseTime on a Token factory with daily capacity in the tens of trillions, and with Wangsu on inference optimization and intelligent scheduling built on Wangsu's globally distributed edge resources.

Zoom out and the acceleration is easy to trace. Qujing Technology was founded in December 2023 and raised a seed round from Zhenzhi Venture Capital in August 2024, followed by an angel round of tens of millions of yuan in February 2025, led by GL Ventures and the Zhipu Z Fund with the Shuimu Tsinghua Alumni Seed Fund participating. The pace picked up in 2026: an angel++ round from Paratera in February, a Pre-A of several hundred million yuan in May co-led by Xinglian Capital and Huakong Technology with Honghui Fund and others joining and GL Ventures doubling down, then the July Series A led by Henan Investment Group's Huirong Fund, with the company announcing more than 1 billion yuan raised in six months. October's Series B added more than 1.3 billion yuan, pushing the year's total to 2.4 billion.

More telling than the pace is the shift in the type of shareholder. At the seed and angel stages, almost everyone around Qujing Technology was a market-driven financial investor. By the Series B, the lead positions are filled almost entirely by state and industrial capital. That tracks directly with the nature of the business. An AI Token factory is capital-heavy, and land, power and chip procurement all sit with local governments. Local state capital puts up the money and the resources, Qujing supplies the system and the operations, and the computing assets ultimately get converted into priceable Token capacity. Henan set the template; Hangzhou moved faster. On July 21, Qianjiang Century City signed with Qujing Technology to put the company's East China headquarters in Xiaoshan, with plans to build an AI Token factory at 10,000-chip cluster scale within five years.

AI Coding Opened the Demand Floodgates

The Token market in 2023 was still early, and suppliers generally chased Token volume over service quality, scaling up supply through brute expansion. Large models were still in pilot and validation mode, not yet embedded deep in enterprise core workflows, so businesses had wide tolerance for shortfalls in performance, stability and model quality. That model sold.

Things changed after Chinese New Year in 2026. AI Coding was the first use case to work at scale: programmers using large models to write code, debug and read entire repositories generate enormous context per call, with a single request reaching hundreds of thousands of Tokens.

Qujing Technology founder Ai Zhiyuan dates the shift to February as well. As the coding capabilities of leading domestic models kept improving, customers moved from being "hesitant to use it" to putting it into production, and "the demand floodgates opened." He compares it to a company leasing a dedicated line rather than buying home broadband: "A dedicated line isn't a luxury, it's a hard performance requirement of a production environment."

He offers one counterintuitive read as well: low prices are not the main thing many enterprise customers are after. "When the price is too low, their first worry isn't whether they got a bargain, it's whether the service can stay stable over the long run."

Two Business Models, the Second One Lighter

Qujing defines high-quality AI Tokens as highly intelligent, low-latency, high-throughput and strongly stable, combined with the ability to deliver at scale at a controllable all-in cost. On the technical side, that translates into "fewer models, deeper optimization, higher quality."

There are two business models. One is leasing computing power: producing AI Tokens in-house and selling them to leading model developers, internet platforms, AI application companies and large enterprise customers. The other serves customers that already have computing capacity or plan to build it, taking on the overall planning and design, construction and delivery of a Token factory plus joint operations afterward, sharing in the ongoing economics once it is running. The second is relatively asset-light, and it is where much of the second half's deal flow has landed.

In June this year, the company's revenue for the single month exceeded its full-year 2025 total; by the latest count, revenue so far this year is close to ten times last year's full-year figure. On the output side, Token production efficiency per unit of computing power has risen more than threefold since Chinese New Year, and daily output of high-quality AI Tokens has grown more than a hundredfold in six months, climbing past the trillion mark. Customers include leading model developers; public information shows Zhipu's GLM and Moonshot AI's Kimi are both served by Qujing Technology.

AI Infra companies are generally stuck growing revenue without growing profit: computing costs come first, paying demand ramps later. Qujing Technology says its overall business has crossed the cost line, and that some of its more mature lines got there earlier.

The Books Balance, Now the Question Is What It's Worth

Token producers are collectively edging toward the public markets. SiliconFlow has already filed with the Hong Kong Stock Exchange, submitting its listing application on June 30, 2026, and by 2025 annual token throughput it is China's largest independent ecosystem token supplier. DataCanvas filed its prospectus on September 29, aiming to be the first AI cloud listing of its kind. Infinigence AI has been reported to have filed confidentially and could list as early as the first half of 2027, which the company has not confirmed.

Policy is opening up too. On June 17, the fifth set of listing standards on the STAR Market was extended to cover large AI model companies. On July 10, the country's first technology innovation corporate bond dedicated to Token computing factories was issued on the Shanghai Stock Exchange.

Ai Zhiyuan's background is unusual among this cohort of founders. A Tsinghua pedigree is not scarce; what sets him apart is that before starting the company he served as R&D lead across several departments at Sangfor, covering big data, digitalization and AI applications, and ran development and delivery on multiple projects worth tens of millions of yuan. He had already gone through the full cycle of research, delivery and scaled deployment inside a listed company. He sums up the transition this way: "In research, what matters most is whether the technology is innovative. In commercialization, you have to look at whether you can supply it, whether you can deliver it, and whether customers can actually use it." He adds, "Stability may not be the most innovative thing in a paper, but in business it can be the top priority, as long as costs stay under control."

The team structure is distinctive as well. Chinese Academy of Engineering academician Zheng Weimin serves as chief science advisor, Tsinghua professor Wu Yongwei as chief scientist, and associate professor Zhang Mingxing is a co-founder and board member. Both CEO Ai Zhiyuan and CTO Chen Xianglin came out of Wu Yongwei's group. In March this year, Wu Wenjie, who holds a PhD in finance from the University of Hong Kong, became president and CFO, filling in the finance and global operations piece.

This round of capital exuberance in AI Infra has left the hardest question, profit, for last. Whether Qujing's answer holds up depends on whether its factory capability can be replicated across more computing centers, more chip combinations and more real customers. That is the test every AI Infra company faces.