CXMT's Shanghai IPO made it China's most valuable listed company overnight. Its own prospectus doesn't fund the one product driving the AI memory story.

ChangXin Memory Technologies opened trading on Shanghai's STAR Market this week at 49.50 yuan a share — up 466% from its 8.66-yuan IPO price — after raising 57.92 billion yuan (about $8.6 billion) in the largest semiconductor listing in the exchange's history and mainland China's second-biggest IPO ever, behind only Agricultural Bank of China's 2010 offering. At the open, that priced the company at roughly 3.31 trillion yuan (about $489 billion), enough to overtake Industrial and Commercial Bank of China as the most valuable company listed on a mainland Chinese exchange. By midday the stock had gone further, to a 531% gain and a market cap closer to 3.66 trillion yuan.

CXMT is China's largest DRAM maker and, by unit share, the world's fourth-largest — an estimated 8% of the global DRAM market in 2025, growing to around 9% in the first quarter of 2026 on the back of AI server memory demand that pushed the company's Q1 revenue to 50.8 billion yuan (about $7.5 billion), up more than 700% year-over-year. Samsung, SK Hynix, and Micron still hold the top three spots by a wide margin.

The gap between the story and the filing

The bull case investors bid up this week rests on high-bandwidth memory — the stacked DRAM that sits next to AI accelerators and is currently sold out industry-wide through most of 2026. CXMT has developed HBM-related engineering capability but has no commercial HBM production, and independent estimates put it roughly three years behind Samsung, SK Hynix, and Micron on the technology. Its IPO prospectus makes that gap explicit rather than papering over it: of the 29.5 billion yuan earmarked for named projects, the money goes to production-line upgrades, conventional DRAM technology enhancement, and forward-looking DRAM research — not a single line item is allocated to HBM.

Part of that gap isn't a choice CXMT gets to make on its own timeline. Under Section 5949 of the fiscal year 2023 National Defense Authorization Act, every US federal agency is barred from procuring semiconductor products from CXMT, its subsidiaries, or its successors, effective December 23, 2027 — and separately, CXMT is currently barred from importing the advanced HBM chips it would need to close the technology gap faster. "CXMT plays a critical role in China's AI push, particularly in the face of U.S. export controls," Brookings Institution fellow Kyle Chan told Fortune. The IPO money is arriving years before the company can spend it on the product the market is pricing it for.

The procurement question was already live in Washington before the IPO priced. On July 16, House Select Committee on China chairman John Moolenaar and Democratic Rep. George Whitesides wrote to Commerce Secretary Howard Lutnick urging tighter export-control restrictions on CXMT and fellow Chinese memory maker YMTC, arguing dependence on Chinese memory chips poses "an unacceptable risk to America's national security, economic security and supply chain security" and that purchases by US companies would help fund People's Liberation Army technology development. The letter followed reporting that Apple had been lobbying for approval to use CXMT memory amid a global DRAM shortage — the same AI-driven demand spike that helped make this week's IPO possible in the first place.

Two analyst reads that don't agree

Wall Street's reaction split cleanly. Nomura's Donnie Teng argued the structural case for absorbing the valuation: as long as hyperscaler AI capex keeps driving memory demand, the market can eventually absorb whatever liquidity the IPO pulled out of Chinese equities, regardless of near-term multiple. Morningstar's read is more skeptical on fundamentals — it puts CXMT at roughly a 30% cost-per-bit disadvantage against Samsung and SK Hynix, a gap that has to close before market-share gains translate into comparable margins. Both can be true at once: a stock can be a reasonable structural bet on Chinese AI-memory demand and still be selling at a valuation that assumes a manufacturing cost gap closes faster than the historical pace of DRAM process catch-up would suggest. Neither analyst view is a recommendation to act on, and this isn't one either — it's the shape of the disagreement, not a verdict on it.

The selloff that mixed up NAND and DRAM

The clearest sign of how fast the market moved without checking its work: SanDisk shares fell roughly 11% the day of CXMT's debut and had lost about a third of their value by July 29, as part of a broader memory-sector selloff that also hit Micron and SK Hynix's US-listed shares. Micron and SK Hynix compete directly with CXMT in DRAM, so their declines track a real competitive threat. SanDisk makes NAND flash, a different memory category CXMT doesn't produce at all — its inclusion in the rout looks like sector-wide contagion rather than a reasoned read on competitive exposure. Reconciling the two: the Micron/SK Hynix moves are a defensible market reaction to a new, well-capitalized DRAM competitor; the SanDisk move is the more recent and more specific reporting, and it's the one this piece treats as the more reliable account of what actually happened, precisely because it identifies why the initial reaction doesn't hold up rather than just repeating the percentage decline.