On October 12, 2026, a Reuters report on Moonshot AI cast a shadow over the AI star company as it Sprint ( Sprint here means 'rushes toward' and is kept for contextual flow, though in formal translation might simply be 'prepares for') its Hong Kong Stock Exchange IPO.

According to Reuters, citing sources familiar with the matter, Moonshot AI founders Yang Zhilin and CTO Zhang Yutao are engaged in settlement negotiations with Recurrent AI to resolve a nearly two-year-long legal dispute ahead of the IPO. A lawyer's letter dated March 11, 2026, obtained by Reuters, shows that Recurrent AI has made multiple allegations, including the unauthorized establishment of Moonshot AI, making false representations to investors, and the suspected unauthorized transfer of servers and deletion of important data.

The timing of the incident is particularly sensitive. On October 6, Bloomberg, citing sources familiar with the matter, reported that Moonshot AI had completed its final pre-IPO private funding round, with a valuation of approximately $50 billion. The company plans to list on the Hong Kong Stock Exchange as early as the first quarter of 2027 and has begun engaging with potential institutional investors. A month earlier, Moonshot AI also showcased to investors its growth achievements, with annualized revenue exceeding $1 billion. The commercialization surge brought about by the release of Kimi K3 is becoming an important basis for supporting the next round of capital pricing.

From technical strength and revenue growth to financing progress, Moonshot AI is experiencing one of the best periods since its inception. The capital market has placed increasingly high valuations on Kimi, and the potential costs Moonshot AI may pay for past entrepreneurial arrangements are also rising accordingly.

The compensation currently demanded by Recurrent AI is equivalent to 10.5% of Moonshot AI's original shares. Reuters, citing sources familiar with the matter, estimates that based on Moonshot AI's funding valuation of approximately $30 billion in June this year, the value of this potential compensation could exceed $700 million.

Moonshot AI is undergoing an ironic capital market test: The equity arrangement that once helped Yang Zhilin gain entrepreneurial freedom may now become the most difficult old contract for him to handle on his path to the public market.

I. Yang Zhilin's Two Entrepreneurial Ventures Leave Three Parties in Dispute

The relationship between Moonshot AI and Recurrent AI is far more complex than the typical scenario of a founder leaving to start a new business.

In 2016, Yang Zhilin and Zhang Yutao co-founded Recurrent AI, primarily exploring the application of artificial intelligence in enterprise services. At that time, foundational large models had not yet become the mainstream focus of the capital market, and enterprise AI relied more on specific scenarios, industry data, and project delivery. By the end of 2022, ChatGPT triggered a global technological paradigm shift, prompting Yang Zhilin to decide to refocus on general-purpose large models and establish Moonshot AI in 2023.

Leveraging its long-text processing capabilities, Kimi gained traction among users and gradually expanded from a chat assistant to areas such as reasoning, programming, and agents. Moonshot AI also grew from a newly established startup into a leading Chinese large model company with investments from giants like Tencent, Alibaba, and Meituan.

However, Moonshot AI's founding team and some early resources were closely linked to Recurrent AI, leading to disputes over issues such as the authorizations granted during the spin-off and the returns owed to the old company for the resources provided to the new one.

In December 2024, Yang Zhilin stated in a personal response that as early as February 2023, he had reached an agreement with Recurrent AI CEO Chen Qicong for Moonshot AI to independently establish a new company, with Recurrent AI obtaining a certain proportion of shares free of charge.

Yang Zhilin also claimed that he relinquished half of his shares in Recurrent AI for zero consideration and helped the original company sell some of Moonshot AI's old shares to alleviate cash flow pressures caused by the Silicon Valley Bank incident at the time. He emphasized that Recurrent AI's board of directors had signed relevant resolutions, and both parties had signed agreements to handle equity arrangements, full-time obligation exemptions, and subsequent cooperative relationships.

This explanation constituted Yang Zhilin's primary stance, but some investors in Recurrent AI believed there were flaws in the relevant authorization procedures and interest disclosures. Disagreements persisted over the validity of the board resolutions, shareholder authorization requirements, and the formation process of the relevant arrangements.

The most sensitive figure in this dispute is Zhang Yutong.

In December 2024, Zhu Xiaohu, Managing Partner of GSR Ventures, publicly questioned that Zhang Yutong, who had previously worked at GSR Ventures, had obtained approximately 9 million initial shares of Moonshot AI, accounting for about 14% of the initial shares, through her involvement in Moonshot AI-related arrangements. In contrast, Recurrent AI received an initial share proportion of 9.5%. Zhu Xiaohu argued that Zhang Yutong had not fully disclosed her personal interests to the relevant fund partners and Recurrent AI shareholders and accused her of breaching her fiduciary duties.

In a public response at the time, Yang Zhilin did not deny granting shares to Zhang Yutong but emphasized that the relevant equity was subject to a multi-year vesting mechanism contingent on her continued service and performance. He stated that Zhang Yutong had made significant contributions in business, strategy, and multiple funding rounds, and the equity represented her long-term compensation as a co-founder.

Yicai Media pointed out at the time that while Yang Zhilin's response explained the commercial rationale behind the equity grant, it did not directly address all external queries regarding shareholding disclosures and potential conflicts of interest.

It is important to note that while a shareholding may have strong commercial incentives, it can still spark controversies over authorization procedures and interest disclosures.

For an AI startup highly dependent on top talent and core management teams, offering substantial equity incentives to key individuals is not uncommon. Especially in Moonshot AI's early stages, fundraising capabilities and commercial strategy held significant direct value for a company requiring substantial ongoing investment in research and development.

However, if the relevant individuals are simultaneously involved in interest arrangements with the old company and acquire personal stakes in the new company, disclosure procedures cannot be simply replaced by consensus among the founders. The magnitude of personal contributions can be discussed, the rationale for equity grants can be evaluated, but fiduciary duties and authorization procedures require independent and explicit bases.

Reuters, citing company records, reported that although Yang Zhilin and Zhang Yutao had founded Moonshot AI, they remained directors of Recurrent AI's holding company. The lawyer's letter also revealed that Recurrent AI and some of its investors were considering filing another legal action against Zhang Yutong under Cayman Islands law. Zhang Yutong had resigned from her position as a director of Recurrent AI in 2022, and the allegations of concealed shareholdings remain to be proven.

Thus far, Moonshot AI's historical disputes have involved three core figures: the current CEO, CTO, and president. If the disputes persist into the IPO phase, scrutiny will extend beyond a single compensation claim to the interest arrangements and governance relationships of the company's most senior management.

Moreover, the contentious issues disclosed in this year's lawyer's letter have extended to technical assets. According to Reuters, Recurrent AI alleged that in April 2023, servers storing critical large language model data were unauthorizedly transferred to a Moonshot AI subsidiary; in September 2024, some important data was deleted following instructions from Zhang Yutao.

For AI companies, training data, model weights, research and development code, and computing power environments are all crucial assets. Once their ownership relationships become legally contentious, they may further involve issues such as technological origins, intellectual property usage, and historical asset allocation.

The closer the IPO approaches, the more quantifiable the compensation demands from old shareholders become; similarly, the stronger Moonshot AI's motivation to resolve the dispute swiftly grows. This two-year-long controversy has now entered a phase where both sides are revaluing their bargaining chips.

II. Revenue Growth Too Rapid: $50 Billion Valuation Hinges on Revenue Quality

Over the past few months, Moonshot AI has presented a set of impressive data. According to a Bloomberg report on September 11, Moonshot AI disclosed to investors that its annual recurring revenue (ARR) surged from approximately $300 million in June this year to over $1 billion in August, with plans to further elevate this metric to $2 billion by year-end.

Based on a rough calculation of a $50 billion valuation and $1 billion in ARR, Moonshot AI's valuation is approximately 50 times its annualized revenue. Moonshot AI is rapidly converting the technological advantages of its star AI product into model invocation revenue. However, the more models it sells, the more the market needs to clarify the nature of this business.

ARR is an indicator annualized from recurring revenue at a specific point in time and cannot be directly equated with realized full-year operating revenue, let alone used to infer actual profits. For companies undergoing significant model updates, short-term revenue surges require particular attention to subsequent retention rates. Developer concentration driven by a single model release differs significantly in long-term value from sustained expansion of client usage in production environments.

In July, Huang Zhenxin, head of Moonshot AI's B2B business, revealed that API invocations accounted for 70% of enterprise business revenue. Meanwhile, the company plans to open up sandbox, Harness, and standardized agent capabilities to enterprises through the Kimi Hosted Agent platform, facilitating the model's integration into actual enterprise workflows beyond single invocations.

This represents a reasonable commercialization path. Compared to acquiring users through consumer-side marketing, enterprise developers can integrate the model into existing products and work systems, generating continuous token consumption. As clients' actual usage volumes increase, model service revenue is directly driven upward. Kimi K3's enhanced capabilities in programming, complex reasoning, and long-term tasks provide Moonshot AI with opportunities to enter high-value production scenarios.

However, the API business conceals a commercial relationship easily obscured by rapid growth: clients purchase model capabilities, but Moonshot AI may not necessarily control the final customer relationships.

When enterprises invoke Kimi through development frameworks, cloud service providers, or third-party applications, the entities possessing customer entry points, workflows, and user data may be service providers positioned above the model layer. While model providers can share in the benefits of usage scale growth, their control over customer migration, usage scenarios, and final product pricing is often constrained by cooperative models.

This structural risk becomes more pronounced during periods of rapid model capability iteration. Developers can adjust traffic allocation among different models based on performance, price, latency, and reliability. A model gaining an edge in programming or agent tasks can swiftly attract clients; however, once competing models catch up, clients may reassess invocation costs.

While experiencing rapid revenue growth, Moonshot AI has also exposed another operational constraint: computing power supply.

On July 19, Kimi announced that due to user requests surging well beyond expectations within 48 hours of the K3 launch, existing clusters were nearing their capacity limits. The company decided to temporarily halt new consumer subscriptions, prioritizing resources to safeguard service quality for existing subscribers while pursuing capacity expansion. This does not mean K3 ceased providing services to developers but illustrates that the demand surge driven by the model release once exceeded the company's service capabilities.

Moonshot AI has not abandoned efforts to address the limitations of merely selling model capabilities. On September 10, the company officially launched the Kimi Enterprise Partner 'Lunar Landing Program,' collaborating with enterprises such as Chinasoft International, Kingsoft Cloud, Teamsun, Akan Inc., and AsiaInfo Technologies. Leveraging frontline deployment engineers (FDEs), the initiative aims to drive deeper integration of the model into client business systems. Partners undertake some client development, engineering deployment, and business delivery work, while Moonshot AI provides model capabilities, technical support, and related tools, with some partnerships adopting token revenue-sharing mechanisms.

For Moonshot AI, establishing a vast industry sales and on-site delivery team would require substantial capital and management resources. Opting for partners to handle the final leg of enterprise service delivery reduces organizational expansion costs and allows the company to concentrate its limited R&D resources on foundational models.

Nevertheless, Moonshot AI has not yet publicly released complete financial documents for its IPO, leaving external parties unable to accurately assess its actual revenue recognition scale, profitability, and cost structure.

While ARR is growing rapidly, gross margins and cash burn rates disclosed in the prospectus have yet to catch up with this valuation narrative.

III. From Zhipu to MiniMax and DeepSeek: Kimi Learns to Accept Peer Benchmarking

Another layer of pressure on Moonshot AI's IPO bid comes from its peers.

Rewinding to 2023, only a handful of Chinese startups could independently train cutting-edge foundational models. Companies possessing top-tier research teams, sustained fundraising capabilities, and productization skills held high scarcity value in investors' eyes. At that time, capital was more willing to bet on technological routes and potential market space, with limited mature financial benchmarks for horizontal comparison ( horizontal comparison here means 'horizontal comparison' and is kept for contextual flow; a more formal translation might simply use 'comparison').

However, with Zhipu and MiniMax successively listing on the Hong Kong Stock Exchange in January 2026, Chinese large model startups now have public financial references. While primary market ( primary market here means 'primary market') investors could previously accept broad valuation speculations, after listing, revenue scale, gross margins, R&D expenses, and loss levels will all come under market scrutiny.

This shift holds particular significance for Moonshot AI.

On August 31, Zhipu disclosed that its revenue for the first half of 2026 reached RMB 954 million, up approximately 400% year-on-year, with a loss of around RMB 2.072 billion during the same period. Its open platform and API business generated RMB 825 million in revenue, accounting for 86.5% of the total, becoming the absolute mainstay of the company's revenue; Zhipu's management also disclosed an August ARR of approximately $1.6 billion.

MiniMax's financial report presented another set of data. In the first half of 2026, the company achieved revenue of approximately $117 million, up 283.1% year-on-year, with a gross margin of 17.9% and an adjusted net loss of around $293 million during the same period. Revenue from open platforms and other AI enterprise services rose to 63.4% of the total. MiniMax's management also revealed that the company's August ARR exceeded $800 million.

It is evident that domestic leading large model companies are simultaneously experiencing API revenue expansion, high R&D investment, and delayed profit realization.

For Moonshot AI, valued at $50 billion, the market will find it difficult to assign a completely independent valuation based solely on ARR growth rates. Zhipu has disclosed its revenue structure, and MiniMax has disclosed its gross margins and R&D expenditures. Moonshot AI will need to provide comparable operating data in its IPO filings to explain its differences in scale, efficiency, and sustained growth capabilities.

Meanwhile, these companies are all vying for developers and enterprise clients. Zhipu is expanding its open platform revenue, MiniMax is driving global invocation growth through models and application products, and Moonshot AI is strengthening programming and agent scenarios with K3. Competition among them has extended from mere model rankings to developer ecosystems, reasoning efficiency, and enterprise workflows.

The presence of tech giants further complicates the competitive landscape.

Alibaba can integrate its Qianwen model with Alibaba Cloud's computing power, storage, and enterprise service systems; ByteDance possesses the Doubao product entry point and Volcano Engine; other internet platforms can also embed model capabilities into their businesses. For independent model companies, training competitive models is merely a prerequisite for commercialization; establishing irreplaceable distribution channels is crucial for long-term market positioning.

DeepSeek's open-source route continues to reshape price expectations in the model service market. As more high-performance models open their weights, technically capable large clients can compare using hosted APIs, self-deployments, or cloud provider model services.

While open-sourcing has gained Kimi global developer attention, it has also made it harder for the model service market to establish exclusivity. Moonshot AI needs to strike a balance between expanding its model ecosystem influence and preserving commercial pricing power. The duration of this balance depends on model update speeds, reasoning cost advantages, and the actual migration costs incurred by enterprise clients.

Of course, the commercialization pressure in the large model industry is not unique to the Chinese market.

A Reuters commentary on October 7 cited Anthropic's listing materials, pointing out that while Anthropic achieved approximately US$4.6 billion in revenue in 2025, its computing power and infrastructure-related expenditures reached about US$7.3 billion, with an operating loss of approximately US$8.06 billion. Clearly, even if a company already has a considerable revenue scale, high investments in foundational model R&D and inference infrastructure can still lead to substantial losses.

Moonshot AI is already not lacking in technological achievements or capital pursuit. Currently, its most costly risks are those issues that cannot be quickly resolved through the release of new models.

To conclude

Yang Zhilin's entrepreneurial choice in 2023 has made him one of the most-watched entrepreneurs in China's large model wave. The release of Kimi K3 and the subsequent rapid growth in commercial revenue also demonstrate that the company's technological investments are gradually yielding market returns.

However, technological achievements cannot replace the handling of historical interest relationships. In 2024, Yang Zhilin publicly explained Moonshot AI's spin-off arrangements and expressed confidence that disagreements would be reasonably resolved. Two years have passed, and as the company's valuation has risen to US$50 billion, related controversies persist, with even new allegations regarding technical data emerging.

Although these allegations remain to be proven, their continued existence has already added complexity to the pre-IPO governance efforts. For Yang Zhilin, the longer the unresolved issues from his time at Recurrent AI linger, the harder it becomes for Moonshot AI's growth to naturally diminish their significance.

From a broader industry perspective, Moonshot AI's situation is also quite representative.

Chinese large model startups are accumulating technological capabilities, commercial revenue, and capital valuations at an astonishing pace. However, the maturation of corporate governance systems, cost structures, and long-term business models often requires more time.

Ultimately, how much a company is worth depends on how much of that value can be clearly and stably attributed to itself.

*The featured image and illustrations in the text are sourced from the internet.