Export controls went after China’s chips. They never touched the capital, and Moonshot is reportedly seeking $5bn.

Moonshot AI is considering raising up to $5bn in a Hong Kong initial public offering as soon as this year, Bloomberg reported on Friday. Dong Cao, Julia Fioretti, Pei Li and Luz Ding sourced the account to people familiar with the matter, who asked not to be identified because the information is private.

The Beijing lab has filed confidentially and added Bank of America as overall coordinator alongside sponsor banks China International Capital Corporation, Deutsche Bank and Goldman Sachs, according to the same reporting. TNW has not independently verified the syndicate or the size, and Moonshot has not announced either.

Note that the figure has moved. Reports a day earlier put the target nearer $3bn, so an editor should treat $5bn as the most advanced version of a number that is still being briefed rather than as settled.

The part export controls were built to prevent

Washington restricted advanced chips to slow exactly this. It has not worked in the way the policy assumed, because the constraint was applied to hardware and the money was left alone.

Moonshot trained its flagship model on Nvidia silicon regardless. Kimi K3 was trained on a 20,000-chip cluster supplied through Alibaba’s cloud, with multiple accounts pointing to H200 accelerators.

The result was not a lagging imitation. Moonshot billed K3 as the world’s largest open-weight model at 2.8 trillion parameters when it launched in July.

Hong Kong is now the exit

The listing route matters as much as the model. Zhipu AI and MiniMax debuted on the Hong Kong Stock Exchange on 8 and 9 January, seeking $560m and up to $539m respectively, as Kinling Lo reported for Rest of World.

Both were oversubscribed, and both went public before OpenAI or Anthropic had filed. Chinese labs reached the public markets first, which is not the sequence anyone in Washington sketched out in 2022.

Moonshot would be considerably larger than either. A raise of the size Bloomberg describes would be several times what Zhipu and MiniMax took between them.

The banks are the tell

Look at who is reported to be arranging it. Bank of America, Goldman Sachs and Deutsche Bank are not Chinese institutions, and their presence on a Beijing AI champion’s listing says something about where the wall actually runs.

Export control governs what American companies may sell. It does not govern what American banks may underwrite in Hong Kong, and the fee pool for a $5bn listing is substantial.

The contrast with the Gulf is instructive. Abu Dhabi’s G42 is reported to be weighing a sale of majority ownership to American companies to keep buying chips, while Beijing’s equivalent is selling shares to public investors with Wall Street running the book.

Why the valuation keeps moving

The numbers have been revised upwards all year, which is its own signal. TNW reported a planned listing within six months at a $30bn valuation, and a subsequent double raise chasing $50bn.

Demand is the stated driver, with annual recurring revenue reported at around $300m and subscriptions paused at one point because the company ran short of GPUs. That is an unusual problem for a lab supposedly starved of compute.

Investor appetite is not the same as durable economics, though. Zhipu lost roughly $330m on $27m of revenue in the first half of 2025, and MiniMax lost $512m on $53m over nine months, on the prospectus figures Rest of World cited.

The competitive question this raises

Cheap capable models are the mechanism, not the listing. Moonshot has been undercutting American incumbents heavily on token pricing, which is the same playbook DeepSeek ran.

That pressure now reaches the US listing pipeline. Cheaper Chinese models are already a factor in how OpenAI and Anthropic valuations get argued, and K3’s release contributed to a global tech and semiconductor selloff that revived DeepSeek comparisons.

Order matters here in a way it usually does not. If Chinese labs establish public-market comparables first, American issuers price into a market that has already been shown a cheaper alternative with published weights.

What is not settled

A confidential filing is not a prospectus, and none of the numbers here are audited or public. Timing, size, valuation and syndicate can all change before anything prices, and Moonshot has said nothing on the record.

Political risk cuts both ways too. Washington blacklisted Zhipu in January 2025, restricting its access to US technology, so a Chinese AI listing carries regulatory exposure that a prospectus has to disclose.

The broader point survives all of that. Restricting chips was meant to constrain Chinese AI, and the capital markets have been quietly routing around the constraint for most of a year.