China's AI Chip IPOs Are Pricing Above Their Private Rounds. US Frontier Labs Are Borrowing to Stay Private.
Cloud Leopard is asking ChiNext for 30.35 billion yuan against a 14.27 billion yuan private mark, CXMT rose 450 percent on debut, and Kunlunxin wants $50 billion in Hong Kong, while the US labs push listings to 2027. Here is where AI price discovery is actually happening, and the two prints that will settle it.
Cloud Leopard Intelligence, a Shenzhen maker of DPU chips, updated its ChiNext prospectus this week to raise 30.35 billion yuan, about $4.2 billion, per 36Kr. Its last private round valued the whole company at around 14.27 billion yuan. Tencent owns 19.78 percent.
Read that again. A company is asking public investors for roughly twice what private investors last said it was worth in total. In the United States that sentence would be a joke. In China this year it is a pattern.
## The pattern
In July, CXMT's Shanghai STAR Market debut rose more than 450 percent after an IPO that raised at least $8.6 billion, briefly making the DRAM maker the most valuable company on a mainland exchange. In June, Baidu's chip arm Kunlunxin was reported to be pushing a Hong Kong IPO at a target valuation of about $50 billion, and Baidu's own shares rose more than 8 percent intraday on the report. LimX Dynamics closed a pre-IPO round of about 13.6 billion yuan at a valuation of roughly 15 billion yuan, which is to say it raised close to the whole company's value in one round on the way to a listing (deal record). Kuaishou's Kling AI spin-off comes with a mandated IPO by 2031 written into its round of up to $3 billion.
Even the exception fits. DeepSeek was in talks at a $71 billion valuation with an IPO targeted for 2027, then put the round on hold after remarks attributed to its founder went viral. The listing was the plan. Politics paused it, not demand.
## The American mirror
The US frontier labs are moving the other way. Anthropic is reported to be planning an IPO of up to $100 billion at around a $2 trillion valuation, with Nvidia weighing a $10 billion anchor. But on September 14 the Nikkei fell 518 points on reports that major US AI firms plan to slow frontier development and delay IPOs, and SoftBank led the decline. Our analysis this week covered the mechanism: SoftBank is funding its OpenAI equity with $11.1 billion of high-yield bonds, and the OpenAI listing has moved to 2027.
So the two largest private AI positions in the world are financed with junk paper and private mega-rounds, while a Shenzhen DPU maker and a DRAM fab go to the exchange and get paid more than their private backers thought possible.
## What is being priced
Two readings. The generous one: Chinese investors have few ways to buy domestic AI exposure, the STAR and ChiNext boards exist to list exactly these companies, and the state wants a self-sufficient chip stack, so demand is deep and policy-shaped. The harsh one: a 450 percent first-day rise and a raise at twice the last mark are what a mispriced market looks like from the inside, and the private rounds were the honest number.
Both can be true at once. What matters for anyone tracking AI valuations is where the public test happens first. For Chinese AI silicon it is happening now, at prices above the private rounds. For US frontier models it has been pushed to 2027, and the private marks have no public check at all.
## What to do with this
First, watch Cloud Leopard's final pricing against the 14.27 billion yuan private mark, and whether the full 30.35 billion yuan survives review. That gap is the cleanest live measure of the premium Chinese public markets pay for AI chips over private ones.
Second, treat the Kunlunxin Hong Kong filing as the next real data point, ahead of any US frontier-lab S-1. Hong Kong prices in dollars with international investors, so a $50 billion print there, or a cut to it, tells you whether the premium survives contact with foreign money. The rounds cited here are in our Tencent deal record and the investment tracker.