Multiply out what a 466% first-day gain does to a book: CXMT sold more than $8.6bn of stock in Shanghai in July, and the close marked that same slab at roughly $48.7bn, a difference near $40bn that accrued to whoever held the allocation rather than to the company's cash line [1]. Shanghai's market, as S&P Global Market Intelligence's Ruiying Zhao notes, is heavily driven by retail investors [12], and a bid willing to pay 5.66 times issue is telling you the binding constraint is supply of paper, not appetite for it.
The venue arithmetic points the same way. LSEG puts Hong Kong and Shanghai at over $54bn of IPO and secondary proceeds so far in 2026 against more than $46bn for all of last year, about 17% higher with the year unfinished [9][4]. Against Nasdaq's roughly 55% of global proceeds, that 21% back-solves to a worldwide pool near $257bn, which means SpaceX's $75bn June listing on its own out-raised the two Chinese exchanges combined by some $21bn and took about 29% of the world's IPO money [11][3]. Two exchanges, or rather two exchanges and a captive savings pool, against one deal.
The strongest case against calling any of this overheated is CXMT's revenue, up more than 700% year on year to 50.8bn yuan, about $7.5bn, in the first three months of 2026 on demand for AI chips [8], which makes the entire raise roughly 1.15 times a single quarter of sales [5]. Perris Lee of ION Analytics reads the deal as a testament to China's tech self-sufficiency ambitions [15], and on the cash he has a point: a memory maker compounding at that rate can grow into almost any first-day mark.
This is probably wrong, but I think the pops price enclosure rather than earnings. Mainland exchanges limit foreign buying, which is why parallel Hong Kong listings exist at all [16]; the overseas route takes longer, according to Freshfields' Howie Farn [18]; and tighter scrutiny on both sides has pushed strategically sensitive issuers home [17]. Rationed supply meeting money that cannot leave produces five-baggers more or less regardless of what sits underneath. Unitree is the cleanest read on that, up 460% on day one in August and then down more than 40% from the debut peak by Friday, which still leaves it near 3.4 times its issue price if that peak sat close to the debut close [6]. The drawdown has not yet reached the money.
What would prove me wrong is dull and checkable: the 2026 cohort still trading above its debut closes a year out would mean the books were priced conservatively into demand that was real, and that underwriters, not households, misjudged. Zhao's own test is whether AI sentiment alone is enough, with a durable cycle requiring sustainable revenue and visible margins [13]. Jacob Cooke of WPIC Marketing + Technologies supplies the other half, that the AI investment cycle is absorbing risk appetite that would otherwise have flowed to a company like Shein [14]. Money going into memory fabs and humanoid legs is money not underwriting consumer platforms, and that is how a fast fashion giant takes a mark far below its peak valuation [2] while selling paper into a queue where it is the least interesting name on the screen.
