Questioning AI · How to Solve Profitability Challenges Behind a Valuation of Billions?

Original First Publication | Jinjiao Finance (ID: F-Jinjiao)

Author | Tian Yu

The frenzy around OpenClaw is waning.

A week ago, it was undoubtedly the top trend in the AI community: deployment tutorials flooded social media, on-site installation services charged premiums up to thousands of yuan, and queues for installation even appeared in Shenzhen. However, as safety boundaries were repeatedly scrutinized, this “shrimp farming craze” quickly cooled down, and market sentiment shifted from blind enthusiasm to caution.

Amid the aftershocks of this hype, a company called StepStar successfully made the leap from behind-the-scenes technology to the forefront of capital attention.

In multiple OpenClaw model usage rankings, StepStar’s Step3.5-Flash once topped the charts due to its high preference for calls. It can be said that, though OpenClaw is stepping back, StepStar has leveraged the momentum to stand out.

In fact, early capital bets had already been placed.

According to multiple media reports, this unicorn, founded less than three years ago, is accelerating Pre-IPO financing, planning to submit an application to the Hong Kong Stock Exchange as early as June this year, with a target valuation of $10 billion. If successful, it will follow Zhizhu and MiniMax, securing the third spot in China’s large model sector.

In the early wild days of AI explosion, true winners are those who can not only handle surging traffic but also convert it into solid valuation.

Currently, it seems that StepStar has firmly caught this “massive traffic.”

The Biggest Winner in “Shrimp Farming”?

OpenClaw rose quickly and fell just as fast.

But it left an important signal: AI Agents are transforming from a technical concept into a real demand.

In recent years, large models have been more like dialogue tools: answering questions, writing articles, generating code. But OpenClaw allowed many ordinary users to experience a new possibility: AI is not just about answering questions but can also complete tasks for you.

Whether it’s intelligent reminders across a week’s schedule or automatic email inspections every few hours with styled summaries, OpenClaw’s practicality gave ordinary users their first direct encounter with the power of Agents.

This is also why, even as popularity wanes, developer communities continue to discuss Agents.

And in these discussions, StepStar is frequently mentioned.

The reason is simple: in multiple OpenClaw model usage rankings, StepStar’s Step3.5Flash once surged to the top globally.

This ranking may not be considered an official authority, but it still holds significant reference value. Because it measures not just paper parameters but whether developers are willing to pay real money for tokens and continue calling in real scenarios.

In a sense, this is closer to the real market.

Furthermore, StepStar’s top ranking reflects the most practical variables for Agent deployment: openness, response efficiency, and cost.

First, openness.

In the Agent ecosystem, developers care less about what questions the model can answer and more about whether it can be integrated, modified, and embedded into existing workflows.

Compared to models that only offer API calls with relatively closed underlying capabilities, StepStar further opens pre-training weights, mid-training weights, and the Steptron training framework. Developers get not just a model interface but a set of modifiable infrastructure.

Next, response efficiency.

Step3.5-Flash’s inference speed can reach up to 350 tokens/sec. In chat scenarios, this just means “faster replies”; but in an Agent system, speed affects the overall efficiency of task chains.

Additionally, Step3.5-Flash is open-sourced under the Apache 2.0 license, allowing free commercial use, which directly reduces the trial-and-error costs for developers and enterprises.

It can be said that StepStar’s visibility in the “shrimp farming” craze is not just about “ranking wins,” but about gaining recognition and acceptance in the most practical variables of the Agent ecosystem.

For a large model company preparing to go public, this market-validated recognition is often more valuable than short-term traffic.

Valuation of $10 Billion

If OpenClaw merely made StepStar more visible, then the real spotlight was turned on by capital.

In the large model sector, fundraising ability itself is a form of competitiveness because this is an extremely capital-intensive business.

Compute power, data, R&D, talent, infrastructure—each requires continuous capital investment. Before the business model is fully proven, those who can keep raising funds are the ones who stay in the game.

StepStar was founded in April 2023.

Founder Jiang Daxin previously worked at Microsoft, serving as Microsoft Global Vice President and Chief Scientist of Microsoft Asia Internet Engineering Institute, responsible for expanding Bing search to over 200 countries and regions worldwide.

The team’s technical background is equally impressive.

Chief Scientist Zhang Xiangyu is one of the authors of the classic ResNet paper; CTO Zhu Yibo was a researcher at Microsoft Research, later responsible for ByteDance’s AI Infra, and worked on GPU products at Google Cloud.

Before the technical route is decided, the team’s background often serves as the most direct valuation anchor.

Because of differences in model capabilities, many investors may not understand the technical details, but knowing where the founders come from and what the team has done is immediately recognizable.

Who brings top talent together is more likely to win the capital market’s bet first.

According to Caixin, in December 2024, the company completed a multi-hundred-million-dollar Series B financing round, with participation from Shanghai Guotou Leading Fund, Pudong Venture Capital, Xuhui Capital, Wuxi Liangxi Fund, Xiamen Guomao, and other state-owned investors.

On January 26, 2026, it further completed a staggering over 5 billion RMB Series B+ round, involving institutions such as Shang Guotou Leading Fund, China Life Equity, Pudong Venture Capital, Xuhui Capital, Wuxi Liangxi Fund, Xiamen Guomao, Huachuang Technology, and others, with additional follow-on investments from Tencent, Qiming Venture Partners, Wuyuan Capital, and more.

Post-funding, the company’s management also saw new changes.

StepStar announced that Yi Qi, Chairman of Qianli Technology and co-founder of Megvii, would serve as Chairman. According to Caixin, Yi Qi will oversee overall strategy, especially focusing on the commercialization of AI large models on terminal devices.

More importantly, the capital market continues to act.

According to Caixin, StepStar is conducting Pre-IPO financing in two tranches, with the first pre-money valuation around $4 billion, and the second between $5 billion and $6 billion; plans to file in Hong Kong before June 30, with a cornerstone pricing around $10 billion, aiming for a listing by year-end.

If it goes public smoothly, StepStar will become China’s third listed large model company after Zhizhu and MiniMax.

The performance of these two earlier companies has also provided a clear valuation reference for later entrants.

After Zhizhu and MiniMax went public, their public offerings were oversubscribed by 1,159 and 1,837 times respectively. By the end of February, both saw share price increases of over 360%, with market caps surpassing HKD 230 billion.

In a sense, the Hong Kong market is establishing a new valuation system for Chinese large model companies.

For StepStar, Hong Kong is no longer just a “listing channel” but a direct contest over valuation and premium capabilities.

How Far Is Profitability?

However, the story of large model companies is never just about technology and fundraising.

The real factor that determines whether valuations can stabilize is: Can they make money?

And on this point, the entire industry has yet to give a clear answer.

Although they are all part of the “Big Six Small Tigers” of large models, their paths differ.

MiniMax emphasizes B2C entry, Zhizhu focuses on B2B solutions, while StepStar bets on multimodal capabilities and terminal deployment.

By 2025, StepStar has released multiple multimodal inference models such as Step3, audio editing large model Step-Audio-EditX, open-source audio inference model StepAudioR1, and GUI (Graphical User Interface) model Step-GUI.

By the end of 2025, StepStar had released 29 models, of which 24 are multimodal.

In terminal collaborations, StepStar is among the domestic large model startups with the most partnerships with mobile phone manufacturers.

Honor, OPPO, ZTE, and others have adopted StepStar’s large model capabilities. Vice President Li Jing previously revealed that their models are deployed on millions of phones, with daily cloud calls reaching tens of millions, a more than 20-fold increase since early 2025.

In the automotive sector, it’s another key battlefield.

StepStar’s collaboration with Geely and Qianli Technology to develop intelligent cockpit AgentOS has been installed in models like Geely Galaxy M9. Three months after launch, sales approached 40,000 units, and the model has entered overseas markets.

It is expected that this year, StepStar’s large models will be deployed on over one million vehicles.

The direct benefit of terminal deployment is earlier revenue realization.

According to Caixin, StepStar’s revenue in 2025 was nearly 500 million yuan, expected to reach about 1.2 billion yuan in 2026. For comparison: Zhizhu’s 2024 revenue was 312 million yuan, with 191 million yuan in the first half of 2025; MiniMax’s 2024 revenue was $30.52 million, with over $53.4 million in the first three quarters of 2025.

While revenue growth is impressive, it’s only part of the story. The key to whether large model companies can stabilize their valuations lies in whether these advantages can ultimately translate into profits. Unfortunately, this remains the most difficult question in the entire sector.

Public information shows that StepStar’s specific profit figures have not been disclosed. But looking at the broader industry, a stark reality emerges:

From 2022 to 2024, Zhizhu and MiniMax recorded total net losses of 3.89 billion RMB and 805.8 million USD (about 5.638 billion RMB), respectively;

In the first half of 2025, Zhizhu’s net loss widened to 2.358 billion RMB; in the first three quarters, MiniMax’s net loss also increased to $512 million (about 3.573 billion RMB).

Even leading players find it difficult to achieve profitability through their core business.

Large models are essentially a heavy-capital race. On one side are soaring compute costs, escalating R&D expenses, and long-term investments around top talent, data, and infrastructure; on the other, commercialization is still in early stages—API calls, enterprise solutions, and terminal deployment—all growing but not yet fast enough to offset costs.

More critically, if a company chooses to develop its own foundational model, this profitability pressure is not temporary. Unlike lightweight applications that can quickly switch to a “low investment, quick monetization” model, large models require sustained training, iteration, multimodal expansion, and ecosystem battles—burning money over the long term.

Within this framework, StepStar’s situation is not unique. Like Zhizhu and MiniMax, it insists on training its own foundational models. This is why, despite rapid progress in multimodal deployment, terminal cooperation, revenue growth, and market heat, it remains difficult to stand out in an environment under industry-wide pressure.

The “Lobster List” top, massive funding rounds, and market attention have pushed StepStar to the center of China’s large model narrative. For this young unicorn, rushing to the capital market and seeking premium valuation has become an inevitable outcome.

But the story of the second half of large models is far from over.

Within the valuation framework set by Zhizhu and MiniMax, the early leaders have already established a preliminary “survivor template” through soaring stock prices and logical consistency.

Whether StepStar can catch up will undoubtedly face more stringent tests from the capital market.

References:

Caixin “Large Model Company StepStar Raises Over 5 Billion Yuan, Megvii Founder Yi Qi Appointed Chairman”

Caijing Weekly “At 38, Tsinghua Top Student Takes Over a Large Model Company”

CAS Research “Rumored to Price at $10 Billion, Third Large Model IPO in China”

Author’s note: Personal opinions for reference only