Moonshot’s $700M Founder Fight Is the IPO Risk Nobody Can Code Away
A 2.8-trillion-parameter AI model is impressive. A $700 million argument over who owns the company is the bit investors should lose sleep over.
The most expensive thing in AI is not the GPUs. It is founders who treat ownership like an awkward conversation they can sort out later.
Moonshot AI has built a 2.8-trillion-parameter model, lined up heavyweight backers, and reportedly filed confidentially for a Hong Kong IPO that could raise about US$3 billion. Yet the thing that may matter most is an old-fashioned boardroom fight: a claim tied to more than US$700 million of the company’s original equity.
That is not a product problem. It is a leadership problem. And leadership problems do not disappear because your demo is brilliant.
The fight sitting underneath Moonshot’s IPO
Reuters reported on October 12 that Moonshot executives are in talks to settle a legal dispute with Recurrent.ai and some of its investors before Moonshot’s planned Hong Kong listing.
The dispute centres on Moonshot CEO Yang Zhilin and chief technology officer Zhang Yutao. They co-founded Recurrent.ai in 2016, then established Moonshot in 2023, while remaining directors of Recurrent’s holding company, according to corporate records cited by Reuters.
Recurrent and certain investors have alleged that Yang and Zhang made false representations to Recurrent investors and enriched themselves at Recurrent’s expense. Those are allegations, not findings. Reuters said it could not independently verify the claims; Yang and Zhang did not respond to its requests for comment, while Moonshot declined to comment. Yang has previously denied wrongdoing, and counsel representing Yang, Zhang and Moonshot has said the case lacks legal and factual basis.
That distinction matters. Nobody sensible should pronounce someone guilty because a lawyer fires off a letter.
But here is the equally important distinction: public-market investors do not need a court judgment to decide a governance mess is too hard. They only need uncertainty.
According to Reuters, Recurrent received 9.5% of Moonshot’s original shares when Moonshot was established. Recurrent and its investors are seeking compensation equivalent to 10.5% of Moonshot’s original shares. One source told Reuters that could be worth more than US$700 million based on Moonshot’s US$30 billion valuation in a June funding round.
Whether the ultimate number lands at US$700 million, less, more, or nowhere near it is not the point. The point is that a meaningful slice of ownership is being argued over just as Moonshot wants the market to put a price on the whole company.
That is a rotten time to discover your cap table has emotional baggage.
Why this is bigger than a founder scrap
Moonshot is not some tiny app with a couple of mates arguing over a spreadsheet. It is one of China’s most watched AI companies.
Its Kimi K3 model launched in July with 2.8 trillion total parameters, 104 billion activated parameters, native vision capabilities and a one-million-token context window, according to Moonshot’s technical materials. The company released the model as open weight. Reuters reported at the time that Moonshot described it as the world’s largest open-weight AI system.
The technical achievement is real. It has helped make Moonshot a serious name in a brutal AI race, alongside Chinese rivals and the American giants investors cannot stop talking about.
Reuters reported in September that Moonshot had confidentially filed for a Hong Kong IPO and was targeting roughly US$3 billion in proceeds. It also reported that the company had to unwind its offshore “red-chip” structure and switch to an onshore China domicile before filing for the listing.
Now add a live arbitration in Hong Kong, allegations involving company data and ownership, possible action against Moonshot president Zhang Yutong, and an IPO prospectus that will need to survive scrutiny from bankers, regulators and institutions who are professionally paid to be paranoid.
You can see the issue.
An IPO is not a graduation ceremony for founders. It is a full-body scan of the company. The market asks: Who owns what? Who approved what? Where did the intellectual property come from? What claims could dilute shareholders? What has management disclosed? And, crucially, can this leadership team be trusted with other people’s money once the excitement fades?
If your answers require a footnote the length of a Tolstoy novel, you have a problem.
The overlooked issue: settlement is not the same as trust
Everyone will focus on whether Moonshot reaches a settlement. Fair enough. A settlement could remove a large, noisy obstacle before the IPO.
But operators and investors should not confuse a settlement with a clean bill of health.
A settlement tells you that two parties found a price at which they were prepared to stop fighting. It does not automatically tell you the governance systems were sound, the board asked enough questions, or the original allocation of equity was handled properly.
This is where boards routinely let businesses down. They treat founder disputes as private unpleasantness, when they are actually a test of governance architecture.
Did the board document approvals? Were conflicts formally disclosed? Were related-party dealings properly reviewed? Was intellectual property clearly assigned? Did directors challenge management before the company became valuable enough for everyone to lawyer up?
The answers determine whether a business has a temporary dispute or a repeatable governance defect.
I have seen this in plenty of businesses, though usually at a less ridiculous scale. When money is tight, people say, “We’ll tidy the paperwork later.” When the business works, “later” turns into a fight about who created the value. Suddenly the bloke who was happy to send messages at 1am is producing screenshots through a barrister.
It is not romantic. It is expensive.
Moonshot’s case is particularly instructive because AI companies move at a pace that makes ordinary startup governance look slow. Teams shift. Models evolve. data moves. Engineers jump between entities. Investors chase the next funding round. That speed is commercially useful right up until nobody can clearly explain who authorised the important decisions.
Then speed becomes evidence.
The contrarian take: this is not mainly a China problem
It will be tempting for some investors to file this under “Chinese corporate-governance risk” and move on. That is lazy.
The mechanics are universal. Founders start one venture, build relationships and know-how, spot a bigger opportunity, form a new entity, bring in new capital, and tell themselves that commercial momentum will smooth over the old obligations.
It happens in Silicon Valley. It happens in London. It happens in Sydney. It happens whenever ambition gets ahead of documentation.
The difference is that Moonshot’s numbers are too big to hide the lesson. A claim attached to 10.5% of original shares is not a side issue when the business is valued in the tens of billions. It is central to the question of who really owns the upside.
And this is why the usual founder worship needs a reality check. Great founders are not excused from governance because they are technically gifted. In fact, the more exceptional the founder and the bigger the valuation, the more disciplined the board needs to be.
A founder can build the product. A board has to protect the company from the founder’s blind spots, incentives and history. If it cannot do that, it is just a fan club with meeting minutes.
What this means for you
You do not need a US$30 billion valuation or a 2.8-trillion-parameter model to use this lesson tomorrow.
First, treat equity as seriously as cash. Keep a current cap table, written grants, signed IP assignments and a clear record of who approved what. If a sensible outsider cannot understand ownership in an afternoon, fix it.
Second, separate friendship from governance. A mate can be a co-founder, director, adviser or investor. But each role carries different rights and duties. Write them down before money arrives, not after.
Third, make your board earn its lunch. Put conflicts, related-party arrangements, new entities, IP transfers and senior exits into formal board papers. If a director does not want it recorded, that is generally a clue it needs recording.
Fourth, before any funding round, acquisition or sale process, run your own diligence. Ask the ugly questions first: Who might claim equity? Where did the code, customer data and trademarks originate? Are there former colleagues who believe they were cut out? What verbal promises have been made?
Finally, remember that governance is not corporate wallpaper. It is an asset.
Moonshot may settle this dispute and complete a successful IPO. It may go on to be a giant. The company’s technology has plainly earned serious attention.
But the lesson is already clear: you can build a machine with 2.8 trillion parameters and still fail the basic management test of making ownership unambiguous.
Do not be impressed by complexity when the fundamentals are messy. The best operators make the hard conversations early, put the answers in writing, and leave fewer landmines for the future version of themselves to step on.