No longer waiting for OpenAI: Why is the Dark Side of the Moon changing tactics now?

(The author of this article is New Position Pro, published with permission from Titanium Media)

Text | New Position Pro

Three months ago, Yang Zhilin wrote in a leaked internal letter: the company’s primary market financing capability is strong, and there is no urgency to go public in the short term.

The tone of that letter was calm and assured, like someone confident announcing a boundary to the outside world. At the same time, he directly articulated the company’s core goal: the most important aim moving forward is to surpass leading companies like Anthropic and become the world’s leading AGI company. This statement aligns with the letter’s calm tone of “no urgency to go public,” clearly indicating that Kimi is focused on the technological frontier, not the capital market.

However, according to a report by Bloomberg yesterday, Moonlight has initiated preliminary discussions with CICC and Goldman Sachs regarding a potential IPO in Hong Kong. The “boundary” drawn by Yang Zhilin three months ago has been pragmatically pushed back in the face of the tidal wave of AI valuations in the Hong Kong stock market.

What has driven Moonlight to this step is the entire Hong Kong stock market, which has entered a rare emotional peak for AI assets in early 2026. In March, AI large model companies, dubbed “lobster concept stocks” by the market, experienced an unprecedented surge. On March 18, MiniMax’s stock price briefly surpassed 1,330 HKD, while Zhiyu set a historical high of 748 HKD. Zhiyu surged 20% that day, with a market capitalization climbing to 333.5 billion HKD, firmly holding the top tier of Hong Kong’s AI sector.

The spread of enthusiasm even exceeded pure AI companies, affecting the entire tech giant sector. The day after Tencent launched the OpenClaw-compatible AI agent WorkBuddy, its stock price jumped 7.27%, with trading volume exceeding 30 billion HKD. These narratives intertwined, collectively pushing the valuation center of Hong Kong’s AI sector to an unusually high level.

This is the best moment and also the hardest to replicate. From Kimi’s breakthrough with long texts to the strategic positioning of AI Coding, and the revenue singularity brought by K2.5, Moonlight has played a hand that is not perfect, but it fits the moment—an elegant situation.

The sentiment in Hong Kong’s AI sector is at a recent high, with a precedent for the VIE route, overflowing enthusiasm in the primary market, and generous pricing in the secondary market. The probability of these four conditions occurring simultaneously is historically uncommon. And peaks do not remain permanently; the tidal forces of valuation have their own gravitational laws, and they will not rise higher simply by waiting.

When Hong Kong Sets the Coordinates for China’s AI

Entering 2026, the rhythm of Moonlight and the entire AI industry has undergone tremendous changes in the past three months. First, there has been a skyrocketing valuation. From the end of last year to today, in several rounds of financing, Moonlight’s valuation has rapidly risen from approximately 4 billion USD to 18 billion USD. Such a significant change in primary market valuation usually occurs under two simultaneous conditions: a business inflection point and an open market window. Moonlight happens to meet both.

In January, Zhiyu and MiniMax listed on the Hong Kong stock market, becoming the first and second Chinese large model stocks, establishing a true foundation for the narrative of AI in Hong Kong. Since going public, the stock prices of these two companies have increased by over 400% this year, far exceeding industry expectations. MiniMax raised approximately 619 million USD in its IPO, with its stock price soaring by 78% on its first day; at its peak, its market capitalization was close to 400 billion HKD.

Prior to this, Chinese large models were in a state of “valuation vacuum”: they could neither adopt the private equity model based on Silicon Valley liquidity like Anthropic nor find comparable valuation logic in the A-share market. The performance of Zhiyu and MiniMax in the secondary market has confirmed the structural pricing power for these “Chinese-style AI assets” for the first time.

This pioneering demonstration effect is not uncommon in the history of capital markets. Google’s IPO in 2004 set the pricing for the search advertising model; thereafter, every large internet company’s listing has, to some extent, used Google as a coordinate. Around 2018, Meituan’s IPO in Hong Kong also provided a new valuation template for the local life service track.

The macro market backdrop cannot be overlooked either. In the first quarter of 2026, the IPO fundraising scale in Hong Kong was approximately 11.6 billion USD, with a year-on-year growth rate of 385%. The core narrative of this recovery is technology and AI. The “application landing” wave ignited by agents, the rise of embodied intelligence, and the increasing influence of Chinese AI in the global developer ecosystem have all contributed to the extremely optimistic fundamental understanding of this sector among Hong Kong investors.

In such an atmosphere, the capital’s qualitative assessment of China’s leading large model companies is undergoing an important cognitive leap: they are beginning to be regarded as core assets genuinely qualified to occupy a position in the next stage of China’s AI competitive landscape. This qualitative change directly drives up the valuation multiples that the market is willing to offer.

Therefore, for Moonlight, completing an IPO within the current window can lock in the current high premium, converting the accumulated book valuation from the primary market into real liquidity and market value endorsement, while under the oversight of the secondary market, forcing itself to establish a more transparent and sustainable financial structure.

This external stimulus also has a more operationally significant dimension. There was some uncertainty regarding whether the VIE route could be smoothly navigated in the Hong Kong stock market. However, with Zhiyu and MiniMax successfully going public, the feasibility of this channel has been validated by the market. Moonlight has completed the VIE structure in its rounds of financing, meaning it no longer needs to incur the costs of exploring from scratch.

Meanwhile, the enthusiasm in the primary market has not cooled down due to the emergence of secondary market options; instead, it has created a resonating effect. According to Z finance, the current round of Kimi’s financing has become frantic, with many institutions willing to pay upfront just to secure an entry ticket. Leap Star, another member of the AI “six unicorns,” reportedly has also surpassed a valuation of 10 billion USD.

The entire sector is experiencing a collective elevation in the name of AI.

MAU Dropped 70%, but Moonlight is Worth More

The external market environment provides the “timing condition” for the IPO, while Moonlight’s internal business inflection point serves as the internal support for Yang Zhilin’s team’s decision to change course. Without this internal support, even the best market window is merely a mirage.

In October 2023, most domestic large model players were still competing on parameter scale and the number of industry cases, while Kimi’s intelligent assistant tapped into a differentiated area largely overlooked by most competitors: long text processing capabilities. The world’s first intelligent assistant capable of handling 200,000 Chinese characters quickly resonated with university students, researchers, and knowledge workers.

This was Kimi’s first highlight moment and established its early product persona.

However, entering 2025, the speed of evolution in the competitive landscape was far more brutal than anticipated. DeepSeek sparked a global technological shock, while Byte’s Doubao and Tencent’s Yuanbao aggressively expanded their territories using massive traffic entry and marketing strategies, leading Moonlight to proactively withdraw from the costly traffic battlefield. After halting large-scale traffic investments, relying purely on natural traffic, Kimi’s MAU plummeted from a peak of 36 million in Q4 2024 to just 9.67 million in Q3 2025.

At the most critical moment, Moonlight chose to proactively retract. It cut off consumer-end entertainment product lines like Ohai and Noisee, concentrating resources on foundational models and agent development. This decision inevitably required enduring pressure at the execution level; abandoning products with an existing user base meant short-term traffic loss and publicly correcting previous strategic paths.

Yet, this strategic gamble on “scale” versus “positioning” ultimately propelled Moonlight to an unexpected yet logical intersection, allowing it to precisely hit the pulse of the era amid the subsequent surge of intelligent agents.

From late 2025 to early 2026, the AI application paradigm centered around agents achieved a leap from conceptual validation to practical deployment within the global developer community. A large number of developers began building intelligent agent systems capable of autonomous planning, multi-step execution, and tool invocation. This working mode places extremely high demands on the long-context processing, instruction compliance stability, and multi-agent collaborative scheduling capabilities of large models. This is precisely the technical core of K2.5: scheduling hundreds of agents to execute tasks in parallel, with task steps reaching up to 1,500.

On OpenRouter, the token consumption ratio of Chinese models has exceeded 60%, and Kimi K2.5 ranks seventh globally on the latest monthly leaderboard with a token consumption of 30 trillion.

At the same time, Moonlight’s commercialization structure is also changing. After a long exploration of commercialization, it finally found a revenue model highly compatible with its technical strengths: API calls, especially from agent developers and overseas tech teams. Less than a month after the release of K2.5, Moonlight experienced explosive revenue growth, with revenues in 20 days exceeding the total for the entire year of 2025.

The appearance of the founder on stage often serves as the most direct annotation of a strategic shift. At the AGI-NEXT summit earlier this year, Yang Zhilin shared the stage with several fellow founders, attempting to reshape external perceptions of the competitive dimensions of large models. Soon after, he appeared at NVIDIA’s GTC site as a guest of Jensen Huang and systematically disclosed the technical roadmap behind K2.5 for the first time. This high-frequency and high-profile global presence, beneath the surface of technical promotion, was broadly interpreted by the capital market as a prelude to the IPO narrative in Hong Kong.

This also marks a personal transformation for Yang Zhilin. When he is laughing and chatting with Jensen Huang at GTC or confirming financing amounts at the negotiation table with CICC, the rock musician who once strummed the strings in the band Splay, trying to deconstruct the world with the melancholy of Radiohead, is evolving into a mature helmsman who understands the laws of capital gravity and can accurately capture market windows.

He still believes technology is the destination, but he now understands that the fuel to reach that destination needs to be sourced from the capital market.

In Conclusion

Furthermore, the discussions surrounding Moonlight’s IPO have an aspect that has been underestimated in public reports: talent retention.

In the field of AI foundational models, talent is the most scarce and hardest to replace strategic asset. The researchers and engineers at Moonlight are currently seen as high-value talent labeled “those who have actually worked on real foundational models,” and there is even a subtle saying in the industry: if you look for the most competent people in domestic foundational models today, DeepSeek and Kimi are often the first two names mentioned. This saying may not be formally certified, but rather reflects industry reputation.

In the AI recruitment market, reputation is the circulating currency. Major companies like Byte and Tencent are aggressively expanding their reserves of AI talent with high compensation packages, and the market is offering increasingly higher premiums for “those who have truly worked on foundational models.” The lack of liquidity in options is becoming increasingly fragile, and Kimi’s core team is at the center of this talent competition.

The public initiation of the IPO process is the strongest commitment Moonlight can make to its core team. The liquidity of stocks in the public market is a prerequisite for the true realization of option incentive plans. In the primary market, the book value of options can be continually pushed higher with each round of financing, but without a secondary market providing an exit path, that book wealth is merely a string of cold numbers.

However, as all of this begins to take shape, the variables in the entire AI industry are also accumulating at an astonishing rate. The explosion of the agent form has compressed the iteration cycle of large model commercialization from “years” to “months” or even “days.” Moonlight’s 20-day income singularity itself is a signal: the few companies remaining at the foundational model table have entered a dynamic race, where any generational shift in technology could quickly reshuffle income structures and market shares.

Pink Floyd illustrated a white light passing through a prism on the cover of “The Dark Side of the Moon,” which disperses into a complete rainbow spectrum on the other side. This imagery unexpectedly fits Moonlight at this moment: whether that light can remain intact after refraction depends not only on the purity of the technological foundation but also on whether, in each upcoming quarter, Moonlight can forge this 20-day “income singularity” into a financial backbone that continues to rise amid complex regulations and fluctuating expectations.

When the bell finally tolls, the market will not remember that you were once a rock musician; it will only count the chips you hold that can realize the future.