Moonshot AI Filed for a $50 Billion Hong Kong IPO. Six Days Later, Anthropic Said Its Chatbot Was Quietly Answering With Claude.
By Michele De Filippo
16 Sep 2026

The pitch investors are being sold

Moonshot AI confidentially filed for a Hong Kong IPO around September 3-4, seeking to raise somewhere between $3 billion and $5 billion depending on the report, at a pre-money valuation north of $50 billion 2 5. That figure is more than ten times the roughly $4.3 billion the Beijing-based company was worth in its 2025 Series C round, and it comes on top of over $5.5 billion already raised from backers including Alibaba, Tencent, Meituan, IDG Capital, HSG, 5Y Capital and China Mobile 2 5 7. Listing as early as the fourth quarter of 2026 would make Moonshot one of the largest Chinese tech debuts on the exchange this year, riding the same wave of investor appetite for mainland AI names that has driven blowout demand for recent Hong Kong and Star Market listings.

The growth story behind the number

What is meant to justify that multiple is revenue velocity. Moonshot has told investors its annualized recurring revenue jumped from roughly $300 million in June to more than $1 billion in August, and it is now targeting close to $2 billion annualized by year end 1 4. The engine is Kimi K3, the large model Moonshot released in July that climbed to the top of several public benchmark leaderboards while reportedly costing a fraction of what comparable American models charge to run 1 4. That combination, frontier-level performance at a steep discount, is the same argument DeepSeek made in early 2025, and it is central to the case that Chinese labs can out-compete on unit economics even while facing chip export restrictions.

Six days into the roadshow, a fraud allegation

On September 10, Anthropic published a threat-intelligence report alleging that Moonshot, alongside DeepSeek, had secretly routed a large volume of Kimi user requests to Claude and served the results back to users as Kimi's own output 3 6. Anthropic said it identified close to 300,000 requests forwarded in this way over a ten-day window, the large majority directed to its Opus model, using 5,380 accounts it describes as fraudulently created 3 6. Because Anthropic blocks direct access to Claude from within China, the accounts were registered to appear as though they originated in Singapore and Japan 3 6. Separately, Anthropic said Moonshot ran what it called cross-session replay, capturing and reusing Claude's full reasoning traces to train its own models, and logged more than 23 million such distilled interactions between May and July, the very window in which Kimi K3 was being developed 3 6. Anthropic added that some of the forwarded prompts contained internal corporate source code and live credentials belonging to real users, and that it could not confirm whether those users had been told their data might reach a third-party model provider 3 6.

Moonshot's muted response

As of the initial Bloomberg and South China Morning Post reporting, Moonshot had not issued a direct, on-record rebuttal addressing the routing allegation itself 3 6. The company's public statements in the days that followed instead focused on a separate matter, dismissing online rumors about its founder as, in its words, fabricated and malicious, and saying it had filed police reports and would pursue legal action over those claims. That pattern echoes July, when a Moonshot enterprise executive denied earlier, narrower distillation accusations in comments to state media without addressing specifics 7.

Why the timing matters for the listing

The sequence is the story here. Moonshot filed for its IPO in the first days of September, Anthropic's allegation landed six days later, and reports of the company's tripled revenue and $2 billion target followed the day after that 1 2 3 4. For a listing built on the premise that Kimi K3 represents genuine, low-cost Chinese frontier capability, an allegation that some of its most demanding responses were quietly Claude's own output is not a peripheral reputational problem. It goes to the core diligence question Hong Kong Exchange sponsors and cornerstone investors will need answered before pricing: how much of the benchmark performance underpinning a $50 billion valuation was actually produced by Moonshot's own model, and how much of the revenue ramp reflects usage of technology the company does not own. Underwriters typically build in weeks of legal and disclosure review for allegations of this kind, and any requirement to address it in a prospectus risk-factor section could push the earliest listing date past the fourth-quarter target investors have been told to expect.

The wider read for Asian AI investors

Moonshot is not an isolated case. Anthropic named DeepSeek in the same report, and distillation accusations against Chinese labs have recurred through 2026 as US frontier developers tighten access controls on their own outputs 3 6. For investors weighing exposure to China's AI IPO wave, the practical takeaway is that headline cost and benchmark advantages now come with a due-diligence layer that did not exist for prior hardware-centric China tech listings: verifying how a model's stated performance was actually achieved. Alibaba, Tencent and Meituan have already committed capital to Moonshot at valuations set before this allegation surfaced; whether new IPO investors demand a discount, additional disclosure, or simply look past it will be an early test of how much scrutiny China's AI champions can expect to face as they tap public markets.

Follow signals beyond the surface.
Learn how Midas turns market change into intelligence.