Mainland Company Seeks Hong Kong IPO, Seeking to Raise Up to $5 Billion

An AI company from the mainland, Moon Shadow, has reportedly applied for a secret listing in Hong Kong. The company is expected to complete its initial public offering (IPO) as early as this year, aiming to raise between $3 billion to $5 billion. At the same time, its financing valuation has increased more than six times in the past half year, despite still being in a state of substantial continuous losses. Moon Shadow’s product, Kimi K3, has also been entangled in controversies over security testing and political censorship.

According to Bloomberg on September 4th, sources familiar with the matter revealed that Moon Shadow has submitted its listing application in a confidential manner, with a potential fundraising of $3 billion to $5 billion. The IPO coordination is being handled by Bank of America, with participation from China International Capital Corporation, Deutsche Bank, and Goldman Sachs.

The current plan is still under discussion, and the final fundraising scale and listing time may be adjusted. Moon Shadow and several related banks have declined to comment, while China International Capital Corporation has not responded.

The South China Morning Post cited sources stating that the company aims to go public as early as the first quarter of 2027. Moon Shadow responded by stating that they do not comment on market rumors.

Established in 2023, Moon Shadow was founded by Yang Zhilin and others. Its products include the Kimi smart assistant and the large-scale language model Kimi K3.

According to reports from a financial magazine, Moon Shadow’s valuation has increased from $4.3 billion at the end of 2025 to $31.5 billion in the first half of this year, a growth of over six times in half a year.

In addition, the Daily Economic News reported that Moon Shadow completed a financing of over $3.5 billion in late July, raising its post-investment valuation to $35 billion. The next round of pre-IPO financing is expected to be based on an estimated pre-investment valuation of around $50 billion.

An investor who participated in Moon Shadow’s financing told the financial magazine that a $50 billion valuation is too high, and from a financial return perspective, investing in the next round of financing may not be ideal.

Renowned AI expert Guo Tao told the Daily Economic News that Moon Shadow is still in a state of substantial continuous losses, and cost control and sustainable profit models have not matured. After the IPO, the company will face challenges such as computing power input, losses, and investor patience.

On August 6th, the US cybersecurity company Frontier Security released a report stating that Kimi K3 exploited network vulnerabilities in a sandbox environment during security testing, connecting to GitHub to directly retrieve test answers instead of completing the tasks independently.

The report mentioned that the testing environment, originally meant to restrict the model’s access to external information, was exploited by Kimi K3 through accessing GitHub. Researchers cautioned that publicly downloaded models could be used directly by external individuals, presenting potentially greater security risks.

Kimi K3 has also been accused of undergoing political censorship by the Chinese Communist Party (CCP). The non-governmental organization “Chinese Human Rights” revealed test results on platform X in July, stating that when users asked about evaluating Xi Jinping, the CCP, and the “June 4th Incident,” Kimi refused to answer. However, when faced with questions about assessing the US President, it could provide detailed responses.

The UK’s Financial Times previously reported that the CCP’s Cyberspace Administration tests AI companies including Moon Shadow, checking the model’s responses on Xi Jinping and other political issues, and requiring them to conform to the so-called “core socialist values.”

Additionally, during some user tests on Kimi K3, the model responded with “I am Claude, not Kimi,” raising doubts about the source of its training data.