Uncovering the behind-the-scenes capital driving force for the largest financing of the "Six Little AI Tigers"
An AI large model financing deal exceeding 5 billion yuan
On January 26, 2026, a financing deal valued at over 5 billion yuan refreshed the record for the largest single-round financing amount of large model startups in the past 12 months — not long ago, on the last day of 2025, large model startup Moonshot AI just officially announced a new round of financing of 500 million US dollars.
The company that set this record is StepFun, one of the few large model startups in the field.
According to the intelligence outlet "Intelligent Emergence", its recently completed over 5 billion yuan Series B+ round of financing involves industrial investors including SITI Pioneer Fund, China Life Equity Investment, Pudong Venture Capital, Xuhui Capital, Wuxi Liangxi Fund, Xiamen ITG, and Huaqin Technology, while old shareholders including Tencent, Qiming Venture Partners, and 5Y Capital further participated in follow-on investments.
"The sheer number of shareholders and the scale of the lineup are extremely rare," a person familiar with the matter commented. He told "Intelligent Emergence" that it only took half a year from the launch of the Series B+ financing to the completion of the whole process.
Following the Hong Kong IPOs of Zhipu AI and MiniMax, and the official financing announcement of Moonshot AI, StepFun continued the capital myth of the large model track at the start of the year.
This company, which has always been known for its low profile and sense of mystery, has begun to fully show its edge — in addition to the high-profile financing amount, another important move of StepFun is the official announcement that Yin Qi will take the helm of the company.
Yin Qi, Chairman of Qianli Technology & StepFun
Previously, Yin Qi, Chairman of Qianli Technology, appeared in StepFun's circle of contacts as a strategic partner in the automotive sector. Now, he has come to the forefront as "Chairman of StepFun", the meaning of which is self-evident —
StepFun has got a spokesperson with a more prominent reputation in the technology circle, capital circle and industrial chain circle; Yin Qi's full control over the direction of StepFun also means that the technical system, operational experience and commercial resources he accumulated at Megvii and Qianli will serve as ammunition for StepFun to win the final battle of large models.
With the listing of Zhipu AI and MiniMax, there was once a widespread argument in the market: the primary market can no longer support the financing of large model startups, and listing will become the watershed between success and failure.
But obviously, the successive large-scale financing obtained by Moonshot AI and StepFun has released an optimistic signal to both the primary market and startups: the primary market still has the strength to support the development of large models; foundational models are still evolving, and investors are still willing to bet on them.
At the same time, the aggressive entry of ByteDance, Alibaba and Tencent, as well as the pending new model of DeepSeek, have also made the primary market's investment criteria for large models increasingly stringent.
A practitioner told "Intelligent Emergence" that the biggest consideration of the primary market for large model startups has become whether independent large models have living space, whether the commercialization logic is unique, and whether the commercialization quality is competitive.
At this point in time, StepFun's new financing valued at over 5 billion RMB is particularly precious.
To understand this round of financing, we can look at the two dimensions that the primary market values most now: technology and commercialization.
First, let's look at technology. Few people realize that StepFun is the only company among the "Six Little Tigers" that truly focuses on multimodality.
Then look at business. The uniqueness of StepFun lies in its entry into terminals in the physical world.
A senior executive of StepFun described to "Intelligent Emergence" that others want to be "China's OpenAI" or "China's Anthropic", StepFun is more like "China's xAI + Tesla".
The unique ecological niche in the large model track has also created a stable capital structure behind StepFun: the diversity and balance of the three parties: US dollar capital, state-owned capital, and industrial capital.
Among the six large model startups known as the "Six Little Tigers", StepFun appeared the latest and maintained a low profile. Interestingly, StepFun, which is collectively backed by the three types of capital, has always been a company with a non-consensus temperament in the large model track.
Amid the consensus, StepFun chose to be the "non-consensus" player
StepFun, founded in April 2023, did not actually appear in the industry until nearly a year later.
In the recollection of a person familiar with the matter, StepFun showed up "late", "so late that the other several large model startups had almost attracted all the attention of the primary market".
At that time, the hottest target in the market was "Light Year Beyond" founded by Wang Huiwen, who returned to the industry — the co-founder of Meituan came back and immediately attracted more than a dozen top institutions and individual investors, which led to a very difficult start for large model startups that emerged after Light Year Beyond. The reason was that Light Year Beyond absorbed a large amount of capital from the primary market.
This also led to the fact that StepFun's first round of financing almost came from the "circle of friends" of several co-founders: HSG, Qiming Venture Partners and IDG.
But in the eyes of many StepFun employees, being "late" is worth it. Jiang Daxin, CEO of StepFun, once mentioned in an interview with "Intelligent Emergence" that StepFun's "lateness" stemmed from thorough preparation:
"We want to accumulate energy to build large models with trillions of parameters", and this energy includes computing power, algorithms, data, and systems that did not get much attention at that time. At that time, StepFun recruited Zhu Yibo, who had a background in ByteDance and had experience in operating 10,000-card clusters, to be responsible for building StepFun's systems.
At the same time, a clear roadmap leading to AGI also took shape in the team's mind: Single modality — Multimodality — Unification of multimodal understanding and generation — World model — AGI (Artificial General Intelligence). Later, no matter how the industry pattern changed, this roadmap as firm as a tattoo never changed.
"At that time, the team had basically filled up all its energy bars," said an early member. "The market will judge the team that is most capable of building large models." Even at a financing time point that was not so favorable, StepFun's debut made it the only company among the Six Little Tigers that became a unicorn relying on its first round of financing.
Facts have proved that thorough preparation allowed StepFun to train the large model Step-1 with hundreds of billions of parameters in less than two months, and it succeeded at the first attempt.
This also gave StepFun's second round of financing an optimistic start. It is obvious that the lineup of this round of financing is far larger than the first round, no matter in terms of the number of participants or the diversity of funds: there are US dollar funds represented by 5Y Capital and Shunwei Capital, as well as strategic investments such as Lenovo Capital.
After a turbulent start, the large model track entered a two-year period of rapid progress — consensus and FOMO quickly formed in the breakout of a few companies and products, and then burst quickly along with the foam.
Among them, the most tempting part was the traffic investment war in 2024. With a traffic investment budget of about 100 million yuan per quarter, Kimi from Moonshot AI and Doubao from ByteDance became popular.
Traffic investment brought not only considerable data and expanded market influence, but also capital — in 2024, Moonshot AI once became the company with the fastest financing speed among the Six Little Tigers:
At the beginning of the year, a financing of over 1 billion US dollars led by Alibaba and followed by Lisi Capital raised Moonshot AI's valuation to 2.5 billion US dollars in one go.
After that, Moonshot AI launched a rapid rolling financing in 2024, pushing its valuation to the highest among the Six Little Tigers, all driven by the popularity of its C-end application Kimi.
The sentiment of the primary market was also affected by Alibaba's entry. "At that time, almost all investors felt that the foundational model phase was over, and the next step would be to compete for products," a practitioner recalled.
The Series B financing launched in mid-2024 was, in the eyes of many insiders, the most difficult round for StepFun. "Why not invest in traffic promotion?" a person familiar with the matter revealed that this was the most frequently asked question by investors at that time. Even some members of the team fell into entanglement: should we allocate hundreds of millions of yuan in budget for growth.
Amid the consensus of traffic investment, the person familiar with the matter said, "it is precisely the moment when belief is put to the hardest test". He remembered that StepFun's team did a lot of first-principle thinking about C-end products:
First, C-end products do not have marginal benefits at present — the more users use them, the more losses the company incurs; second, the data generated by C-end users can only be used to build user portraits, which contributes little to actual model training; third, the economic model of traffic investment is not viable, and cannot guarantee product retention.
The final conclusion pointed to prudence in C-end products and traffic investment. "Even if the product data was good, they were very restrained about C-end products," the person familiar with the matter told "Intelligent Emergence". StepFun only used very little resources to verify its products.
At that time, StepFun's internal judgment was still to focus on foundational models, especially multimodal models. To achieve AGI, the model must not only be smart enough, but also be able to "see", "hear", feel and understand everything.
An employee recalled that in 2024, StepFun's core goal was to follow the AGI roadmap determined on the first day of its establishment, and go all out to transition from native multimodality to the integration of multimodal understanding and generation, which is a more difficult roadmap that requires a longer R&D cycle.
This also means that the shareholders who entered in the Series B round also faced the moment of testing their belief. An investment manager participating in the Series B financing said that all the capital parties of the Series B round were attracted by the trillion-parameter language model Step-2 and the multimodal understanding large model Step-1.5V released in July 2024.
These model believers include the newly entered Shanghai State-owned Capital Investment Co., Ltd. and its affiliated funds, Hong Kong's HKIC, Tencent Investment, and some old shareholders who continued to place their bets.
"StepFun takes a steady and step-by-step approach, and is not aggressive enough in all aspects, which leads to missing some time windows," a senior executive of StepFun commented, "but time will prove that steadiness is a precious advantage."
A year later, the Series B+ financing launched in July 2025 became a rich reward for StepFun's adherence to its belief.
During this period, RL (Reinforcement Learning), Coding, Agentic... there were endless hot spots in the market, and FOMO sentiment was everywhere. At the WAIC (World Artificial Intelligence Conference) in July 2025, among a large number of models focusing on Coding and Agentic, the model answer sheet submitted by StepFun looked rather "alternative":
A reasoning model integrated with multimodal capabilities, StepFun's third-generation foundational model Step-3. Thanks to the special design at the system level, the reasoning efficiency of Step-3 reaches 300% of that of the once extremely popular DeepSeek-R1.
Behind this lies another non-consensus choice of StepFun: it is the only startup team that has invested heavily in building self-owned AI Infra. The special emphasis on enabling large-scale application of models with great cost performance through system capabilities also reflects StepFun's pragmatic side.
Multimodality is still StepFun's technical belief, and the only way leading to AGI. "No matter how the industry changes, StepFun has never deviated from the roadmap to AGI since the first day of its establishment," the above executive told "Intelligent Emergence".
This has also brought StepFun a very special shareholder structure: most of the shareholders have only invested in StepFun among the Six Little Tigers, or have made the largest investment in StepFun. "People who love StepFun will love it very much," he concluded.
From the exhibition match to the knockout round
The still fast-growing model track and the unpredictable industry pattern will not give players too much time to catch their breath and indulge in self-satisfaction. After three years of trial and error, the exhibition match decorated with beautiful data is over. To gain recognition from both the primary and secondary markets, model companies must deliver solid performance.
At the start of 2026, capital has made the final choice: with the IPOs of Zhipu AI and MiniMax, and the financing of StepFun and Moonshot AI, these four companies remaining in the final round all have similar cash on their books, all reserving ammunition and polishing their weapons to stay in the game.
For StepFun, it is also time to bid farewell to low profile and show its edge. It can be seen that since 2025, several core technical experts of StepFun have actively come to the front stage to share their insights.
"Most companies only have the CEO as their spokesperson, but StepFun has more than one core figure," a StepFun employee described.
Supporting StepFun's technical capabilities are CEO Jiang Daxin, who once led the R&D of well-known products such as Bing and Microsoft 365 at Microsoft, CTO Zhu Yibo, who once served as the head of AI Infra at ByteDance, and Zhang Xiangyu, core author of ResNet (the most cited paper in the world since the 21st century) and chief scientist of StepFun — an investor once told "Intelligent Emergence" that even if StepFun's core talents start their own separate businesses, they will still be hot targets in the market.
From left to right: Jiang Daxin, CEO of StepFun, Zhu Yibo, CTO of StepFun, and Zhang Xiangyu, Chief Scientist of StepFun
With Yin Qi taking his proper position, StepFun may enter a more determined final-round mode that is better prepared for long-term operations: the team not only has top talents sprinting for AGI, but also has clear-headed helmsmen who have won tough battles and experienced the cold winter of the four AI dragon companies.
Next, commercial performance will be crucial. Now, following the path of OpenAI and Anthropic, most enterprises in the final round have adopted a conservative strategy in business models: Zhipu AI focuses on To B, while MiniMax and Moonshot AI focus on To C and develop software products.
The two most mainstream commercial narratives in the large model track are subscription for C-end users, and API sales and model customization for B-end users. Now, the other three large model startups remaining in the market, Zhipu AI, MiniMax and Moonshot AI, all choose similar commercial paths.
StepFun, which takes the "AI + terminal" path, looks rather different among them.
People familiar with the matter told "Intelligent Emergence" that StepFun made a rational elimination method among the mainstream business models: