Jieyue Xingchen Secures Largest New Year Financing of 5 Billion Yuan: What Does Capital See in It?
At the beginning of 2026, a Series B+ financing round worth up to 5 billion yuan was quietly completed. The recipient was Step Star, an AI foundation model company established less than three years ago. This deal also refreshed the single-round financing record in China's large model track in the past year.
The timing of this financing is quite significant. Throughout 2025, both the amount and the number of financing rounds for Chinese large model companies declined. Valuation contraction, inference cost control, and commercialization verification became common challenges faced by the industry. The concentration effect of funds intensified, and capital became more cautious. The market shifted from pursuing expansion to examining structural efficiency, and the Matthew effect became more prominent.
Against this backdrop, this counter-cyclical financing stands out. On the one hand, it breaks the expectation of capital withdrawal during the "financing winter." On the other hand, it also implies another signal: the industry's reevaluation of "system capabilities," "delivery capabilities," and "scenario embedding capabilities" is becoming the core of the betting logic in the new stage.
This is not just a financing event but may also be a judgment signal for crossing the cycle.
01
After the fever subsides, the criteria that capital values are changing
Regarding this round of financing for Step Star, the time window is worth noting: the entire Chinese large model industry is undergoing a structural cycle transformation.
According to statistics from Sina Finance, Chinese large model layer companies completed a total of 22 financing rounds in 2025, with a total disclosed amount of 9.416 billion yuan. Compared with the stage in 2024 when large-scale financings were frequent and valuations doubled, both the frequency and scale of financing declined. The market entered a transition period from extensive investment to concentrated betting.
Among them, only three companies had a single-round financing amount exceeding 1 billion yuan in 2025: MiniMax, Zhipu, and Dark Side of the Moon. The once "universal capital dividend" is no longer available, and the threshold has risen rapidly. Investors are now more strictly evaluating whether a company has a real product path, self-sustaining ability, and a complete closed-loop for project implementation.
Meanwhile, industry resources are being reallocated. First-tier Internet companies are gradually tightening their external investment windows and returning to building their own model systems. Since 2025, core resources of Baidu, Alibaba, Tencent, etc. have basically focused on building internal R & D systems, and the behavior of "substituting investment for R & D" in the market has significantly shrunk.
In addition, some leading companies in the primary and secondary markets have completed their listings. In the past month, Zhipu and MiniMax successively listed on the Hong Kong Stock Exchange, becoming the first batch of IPO companies in this track. As of the time of publication, the market values of the two companies reached HK$95.5 billion and HK$121 billion respectively, indicating that the capital market still has enthusiasm for pricing leading enterprises.
However, the revenue and profit data reflect another side of the reality:
MiniMax's revenues from 2022 to January - September 2025 were $0, $3.46 million, $30.52 million, and $53.437 million respectively, and the corresponding losses were $73.7 million, $269 million, $465 million, and $512 million respectively. Zhipu's revenues from 2022 to January - June 2025 were 57 million yuan, 125 million yuan, 312 million yuan, and 191 million yuan respectively, and the net profits were -144 million yuan, -788 million yuan, -2.958 billion yuan, and -2.358 billion yuan respectively.
"Technological advancement" has not automatically translated into a "commercial closed-loop" nor solved the problem of "losses." The tension between valuation and delivery is being reexamined. Is the real core ability of a large model company the model itself or the delivery path?
Peng Deyu, a technology industry commentator, told Guanchao Technology Pro: Against this background, the market sentiment has significantly shifted. From the early "parameter scale competition" and "list ranking of capabilities," it has now shifted to the pursuit of "scenario traction" and "product closed-loop capabilities." "Whether it can truly reach users and generate revenue" is replacing "technological leadership" as the core indicator for financing judgment.
Li Rui, the executive director of Qishijie Beijing Technology Co., Ltd., put forward further views to Guanchao Technology Pro: He believes that the deeper change also includes the adjustment of the capital structure. Since 2025, insurance funds, local state-owned assets, and industrial capital have become the main investors after the Series A round. This type of capital prefers targets with clear project paths and the possibility of industrial collaboration, that is, companies with "medium - to long - term construction capabilities."
This does not mean that the industry is entering a "capability - driven deliverable period" from a "traffic - driven technology speculation period." Behind the structural clearance is the end of the narrative dividend and the reevaluation of engineering capabilities.
The companies that truly survive may talk less but have started to do more in - depth work.
02
Why did it receive this "largest counter - cyclical financing"?
When capital is shifting from "extensive investment" to "prudent investment," Step Star completed the largest single - round financing in China's basic large model field in the past year, a Series B+ round exceeding 5 billion yuan. This is not just a capital flow but more like a collective confirmation by the capital market of a certain direction, path, and organizational ability in a cautious atmosphere.
The participants in this round include industrial investors such as Shanghai State - owned Investment Leading Fund, China Life Equity, Pudong Venture Capital, Xuhui Capital, Wuxi Liangxi Fund, Xiamen ITG, and Huaqin Technology. Old shareholders such as Tencent, Qiming, and Wuyuan further followed up. By observing the structure of this financing, it is not difficult to see the underlying logic. It covers insurance funds, local state - owned assets, industrial capital, and market - oriented institutions, showing the characteristics of "distributed + deep pockets."
Specifically, China Life Equity, as a professional private equity investment platform under China Life Insurance and the first private equity investment fund for insurance funds approved for establishment, has always had a strict review process for projects, usually representing the standards for capital safety and cycle tolerance. Local state - owned assets such as Xiamen ITG, Pudong Venture Capital, and Xuhui Capital more reflect the expectation of post - investment resource collaboration and regional industrial binding. Huaqin Technology, as the global leader in smartphone ODM with a market value of nearly 100 billion yuan, appearing in the investment in a large model project, sends a signal that the manufacturing chain is moving closer to the AI - native ecosystem. The continuous additional investment from old shareholders such as Tencent, Qiming, and Wuyuan constitutes a "secondary confirmation" of the company's medium - to long - term development logic. This combination of capital structure is particularly rare in the current cold market sentiment, indicating that capital has shifted from valuation judgment to the evaluation of organizational ability and deliverable paths.
From the perspective of underlying capabilities, Step Star has been continuously building a closed - loop system of "model - engineering - product" in the past two years. As of now, the company has released more than 30 basic models, covering multiple modalities such as language, vision, voice, and 3D images, and has built a unique "1 + 2" basic model system around AI + terminals. Among them, "1" refers to the Step series of base models, which has now been iterated to Step 3, with native multi - modal inference capabilities. According to the disclosed data, the inference efficiency of Step 3 on domestic chip platforms can reach up to 300% of DeepSeek - R1, and chip compatibility and decoding path optimization are considered at the architecture stage, showing a strong sense of deployment engineering. "2" points to two major directions: multi - modal capabilities (text, voice, image) and the combination of terminal and cloud.
In terms of the actual application of AI + terminals, Step Star has carried out cooperation in key terminal scenarios such as mobile phones and cars and has obtained initial market verification. For example, its models have been installed in more than 42 million devices of mainstream brands such as OPPO, Honor, and ZTE, serving nearly 20 million person - times per day. On the smart car side, the AgentOS system developed in cooperation with Geely Group has been installed in the Galaxy M9 model, with nearly 40,000 units sold within three months after its launch at the end of 2025. It is expected that more than one million cars will be equipped with Step's large model this year.
In addition, according to reports from the investment community, as of the end of 2025, the terminal Agent API call volume of Step Star increased by nearly 170% for three consecutive quarters. In the past year, the API call volume of Step Star's open platform increased by nearly 20 times, and the number of active users increased by 5 times. This marks the initial formation of its productization attempt in the To C interaction dimension.
Behind this path is the systematic promotion of "hardware - software collaboration and model - product integration." While most companies in the industry still stay in the role of "service platform providers," Step Star has chosen a more product - oriented path, that is, to build a collaborative ecosystem with terminal manufacturers through its own model system, middleware components, reducing dependence on cloud platforms and enhancing control over implementation.
From the perspective of engineering organization, Step Star is led by a technical team with AI - native backgrounds. The founding members include Zhang Xiangyu, the author of ResNet, Jiang Daxin, the former vice - president of Microsoft, and Zhu Yibo, the former head of ByteDance's AI Infra. They have a trinity of algorithm, system, and deployment capabilities. Meanwhile, according to a report from Phoenix Tech, Yin Qi, the chairman of Qianli Technology, has officially become the chairman of Step Star, an AI startup. Covering terminal scenarios such as smart cars and AI hardware, according to insiders, Yin Qi will stand in the position of an industry integrator to promote in - depth collaboration between the two companies and accelerate the integration of large model capabilities into the physical world.
It is this "structural ability" rather than "parameter ability" that has become a more important evaluation dimension for investors at present. Currently, the relative gap in model performance no longer constitutes a core barrier. Only companies that can penetrate the business process, embed into the physical world, and deliver usable products have the foundation to break out of the valuation game. In this sense, the financing received by Step Star is a collective response from investors to the certainty of its path "from underlying technology to final delivery," rather than a short - term bet on a certain technological performance.
This also explains why, at a time when the valuation of the entire track has generally declined and incremental capital has become more conservative, this financing can still be successfully completed. It is not a continuation of the previous market trend but a confirmation of the direction under the new cycle logic.
03
Model capabilities are no longer scarce. Who can complete the delivery process?
If the core variable in the competition of large models in the past two years was "whether the model is advanced," then in 2026, the importance of this variable is significantly decreasing. With the overall improvement of basic model capabilities and the accelerated spread of open - source and engineering reuse, model performance itself is no longer able to form a long - term moat. The focus of industry competition is shifting from "who can build a stronger model" to "who can truly embed the model into scenarios and form a sustainable product and delivery system."
This change is directly reflected in the differentiation of commercialization paths. The current mainstream models - cloud API calls, To B customized projects, and To C subscription products - have all exposed their respective structural bottlenecks. A practitioner in the large model industry told Guanchao Technology Pro: The marginal revenue of the API model is rapidly compressed under price competition. Customized projects are difficult to replicate on a large scale, and the delivery cost is highly linear with the revenue. C - end subscriptions are highly dependent on traffic and distribution capabilities, which is more in the advantage area of Internet platforms. Against this background, if basic model companies only stay in the role of "capability providers," their business space is being continuously compressed.
Peng Deyu further pointed out: It is under such industry constraints that Step Star has chosen to focus on embedded delivery in terminal scenarios. Mobile phones and cars have become the main carriers for verifying the productization ability of its models, which is essentially a choice with a higher threshold: the models not only need to be "usable" but also "able to run stably, be continuously updated, and integrate into complex systems." Taking the mobile phone side as an example, model installation does not equal a commercial closed - loop. It involves multiple variables such as terminal - side computing power adaptation, system - level calls, interaction logic reconstruction, and the real usage frequency of users. In the car scenario, the models need to meet more stringent safety, reliability, and engineering standards, and the productization difficulty is significantly higher than that of cloud services.
Li Rui believes: From this perspective, mobile phones and cars are not just "traffic entrances" but more like ability verification fields. Whether a model can run stably on terminals, be continuously iterated, and be repeatedly called by real users has become an important standard for judging its productization potential.
The cost of this path choice is also clear. Terminal collaboration means a longer R & D cycle, heavier engineering investment, and more complex industrial collaboration, which also means that the business results will not appear quickly. In a situation where the industry as a whole is still facing profit pressure, this is a path that requires higher organizational ability, capital patience, and execution ability. Therefore, the number of companies that can truly complete this path will not be large.
This may be the watershed in the third - stage competition of large models: after the convergence of model capabilities, what determines a company's long - term value is no longer the parameter scale or list ranking but whether it has a real path to transform the model into a deliverable system. Whether it can "return" from the scenario to form a positive cycle of data, products, and engineering is becoming a new screening mechanism.
In this sense, Step Star's "AI + terminal" is not a conclusion but a choice that is being tested by the market and time. Whether it is valid depends on whether the delivery can be continuously expanded, rather than a single installation or a phased cooperation itself. This is also a problem that the entire industry must face together after entering the deep - water area of productization.
Author: Gao Heng, a member expert of the Science Fiction Communication and Future Industry Special Committee of the China Science and Technology Journalism Society
This article is from the WeChat public account "Guanchao Technology Pro," written by Gao Heng and published by 36Kr with authorization.