Two banks, three of the biggest AI listings on Earth, and one very thin wall between them. What could possibly go wrong? If you're just joining us: Greater China's AI listings were already past the first-pop phase. Zhipu AI and MiniMax surged in Hong Kong, and then the story turned. It stopped being, 'How high do they open?' and became, 'Can the market absorb new supply once lock-ups expire?' Zhipu's roughly four-billion-dollar secondary share sale is the live test of whether public investors keep funding Chinese AI names after the rush cools. This is the one to stay for. Today — a luggage maker accidentally prices DeepSeek, Databricks reprices itself in private, and Goldman quietly serves three rivals at once. Let's start with the banks. Financial Times writes:
Goldman Sachs and Morgan Stanley have secured joint lead status on all three major 2026 tech IPOs: SpaceX, Anthropic, and OpenAI. This dominance creates a containment problem as banks erect Chinese walls to prevent information leakage between competing clients. Separate deal teams with zero personnel overlap have been established, but equity sales forces and research departments remain shared.
Here it is. Goldman and Morgan Stanley, joint lead on SpaceX, Anthropic, and OpenAI — all three. Same building, three clients who'd happily bury each other. This is the thing no S-1 will ever describe. Separate deal teams, zero personnel overlap — sure. But the equity sales force and research? Shared. The FT says it right there. And that shared sales desk works fine as long as demand is bottomless. The whole setup holds because the salespeople are just executing, not choosing. The tell in the FT piece is the conditional — 'if demand proves finite.' And SpaceX's equity and bond volatility already suggests demand might be finite. So now the same desk has to ration investor interest across three rival deals and treat each issuer with, quote, scrupulous fairness. Good luck subpoenaing your way to that. My favorite line in the whole piece — bankers 'might consider deleting WhatsApp.' You can hear the post-mortem in that sentence. Here's Finimize:
The Coatue-led round, which Databricks expects to close later this summer, would bring in new and existing backers; The Wall Street Journal said it could total about $3 billion. Databricks sells a platform that helps companies pull data from many places, analyze it, and use it to build AI tools, putting it in direct competition with Snowflake.
So Databricks signs a term sheet — Coatue-led, roughly three billion, closing later this summer — at a $188 billion valuation. That's a 40 percent step-up from the $134 billion mark earlier this year. And unlike a lot of what we chase on this beat, there's actually a disclosed structure here. A lead investor, a check size the Journal pins near three billion, a term sheet. I can interrogate that. Right, but here's what I keep coming back to — if you can reprice yourself up fifty-four billion dollars in six months without filing a single document, why would you ring the bell at all? Because eventually somebody wants liquidity. But the trap you're building — the next time Databricks sets a price in public, it gets measured against this private mark. A $188 billion floor is a hard number to grow into on day one. And that's the private-market seduction. You never have to open flat. You just print a bigger step-up and call it validation. Here's The Next Web:
Moonshot AI has told investors it is preparing to list in Hong Kong in as soon as six months, Bloomberg reported on Saturday. The company has distributed a shareholder resolution seeking backing for the IPO and is wrapping a fundraising round that may value the three-year-old startup at more than $30 billion. Annual recurring revenue hit $300 million in June, up from $200 million in April.
Add Moonshot to the Greater China listing rebound: it's telling investors a Hong Kong listing could be six months away, at around a $30 billion valuation. The ratio that matters is $300 million of ARR against that $30 billion valuation — a hundred times revenue. Three-year-old company, ARR up from $200 million in April to $300 million in June. That's a real curve. But the valuation is a fundraising round that 'may value' — those are Bloomberg's words, not a prospectus figure. Right, and watch what actually moved the timeline. The article says they were ready to pull the trigger before Kimi K3 dropped — then the model matched Fable 5 and GPT-5.6, tanked global tech stocks, spooked David Sacks, and suddenly the roadshow has a demo no pitch deck could buy. That's boardroom pressure, not a spreadsheet. The document I care about here is the VIE cleanup. They're dismantling the offshore VIE structure, which is the concrete regulatory move under the CSRC's revised rules. That's what turns a Hong Kong listing from aspirational into mechanical. And it answers a question I've been chewing on. Databricks just reprices to $188 billion on a private term sheet — so why ring the bell at all? Moonshot's answer is competitive positioning. You don't stay private when your model just beat the American frontier labs and you can print that on the cover. A hundred-times-ARR bet, priced off a model release two weeks old, listing in a venue chosen partly for regulatory friction. Put that next to whatever Anthropic eventually prints and it's a very interesting comp. Here's Alina Maria Stan at The Next Web:
It surfaced in a stock-exchange filing by Anhui Korrun, a Chinese luggage maker, which discloses a stake small enough to imply the Hangzhou AI startup is valued at around $52bn. Korrun’s subsidiary Ningbo Purun put 2.9bn yuan into a fund managed by Monolith Management, taking a 0.8265% indirect stake in DeepSeek at a level comfortably above the$45bn valuation floated in May.
A suitcase manufacturer. That's who finally gave us a hard DeepSeek number. Anhui Korrun's subsidiary buys a 0.83% indirect stake through a Monolith fund, files it because they have to, and reverse-engineer the arithmetic — 350.88 billion yuan, call it $51.8 billion. And here's what I love about it — for once I can point at the source. We get a stock-exchange filing, first reported by Reuters, with traceable math — no banker leak, no anonymous 'people familiar.' Compare that to the $71 billion figure that was floating around a few days back — no document, no entry point, just vibes. Korrun's filing pins it at $52 billion off one fund's actual check on one actual date. And that's exactly the asymmetry I keep hammering. The most precise DeepSeek valuation we have came from a luggage company's compliance disclosure — nowhere near the lab, its bankers, or its founder. A sophisticated buyer knows how to hunt for that. Retail has no idea it exists. Though let's be honest about what $52 billion actually is — it's an implied number from a 0.83% sliver. It tells you what one fund paid on one date. It drifts the second a bigger check clears. Right, treat it as a data point rather than a full round. But it does sit comfortably above the $45 billion that was floated in May — so the direction's clear even if the decimal isn't. From Madeline Shi at PitchBook:
A handful of recent PE-backed IPOs have done away with the lock-up or at least compressed the timeline. While this makes sense in a market where investors want quick cash, for some, it raises questions about sponsors’ conviction in the businesses they’ve listed and the broader purpose of public markets.
Here's the one I've been waiting for all week. Madeline Shi at PitchBook — sponsors selling down inside the 180-day lock-up. Forgent Power went public in February, and by early July the company and Neos Partners had run three follow-on offerings, all inside the window. And this is PitchBook's language, not mine — the rush 'raises questions about conviction.' When the sponsor who knows the company best is out the door before the lock-up even expires, that's a tell the retail buyer never sees in the prospectus. Yeah, no — and remember what a lock-up is even for. It exists to stabilize the stock in those volatile early months by keeping supply off the market. Neos compressed it three times. At that point, you're basically doing a fire sale with a press release. But I'd push on the word 'conviction.' Returning cash to LPs can be about the clock more than the business. The board's under pressure to distribute, the fund's aging. That timeline lives nowhere in the S-1, and it drives half these decisions. Sure, but the retail buyer can't distinguish 'we lost faith' from 'the fund needs a distribution.' Both look the same in that window. To the person who bought on day one, the effect's identical — the smart money's gone and they're holding. Fair. And 182 million shares moved on Forgent. Whatever the motive, that's the number that matters — the supply the lock-up was supposed to hold back showed up anyway. If AI IPO Watch helps you stay ahead of the market, follow or subscribe wherever you're listening. And if you've got a second, leave a quick review — it really helps other people find the show.
Next, we're watching Databricks' Coatue-led strategic funding round, which is expected to close later this summer. For Moonshot, the key checkpoints are the shareholder resolution and the fundraising close before any Hong Kong IPO filing within six months.
You'll find links to every story we covered in the show notes, if you want to dig into the ones that matter most to you.
That's AI IPO Watch for today. This is a Lantern Podcast.