CXMT's Shanghai Surge, Intel Foundry Losses & Memory's New Supply War

**Jamie Cole** (0:00)
AI Hardware & Chips Daily News I'm Jamie Cole. Thanks for joining me. Today, CXMT's 466% debut market panic or rational supply chain hedging. A Chinese memory chip maker nobody outside the industry had heard of last month just posted a 466% gain on its Shanghai debut and it briefly sent semiconductor stocks across three continents into a tailspin. That's the story worth understanding today. CXMT listed on the Shanghai exchange at a valuation of 3.3 trillion yuan. SK Hynix fell 13% intraday. Samsung fell 13%.
Nvidia dropped 5%.
The sell-off recovered by Friday on the back of strong Amazon and Microsoft results, but the signal was clear. Markets treated CXMT's debut as a threat signal, not a local IPO story. The important distinction is whether that reaction was rational or just noise. Here's the case for rational. The global DRAM shortage is real and worsening. Samsung's CFO has confirmed shortages will intensify through 2027 and into 2028
16GB of RAM that cost the equivalent of 300-400 Hong Kong dollars a year ago now costs 1,500. That's a 5-6 times price spike and it's already feeding into PC and smartphone pricing.
In that context, any credible new supplier looks valuable. Here's the catch. CXMT isn't a credible peer supplier yet. The company carries a 20-30% cost per bit premium over SK Hynix, Samsung and Micron and it sits 2-3 generations behind on process technology. Analysts don't see it closing that gap before 2028 at the earliest. So what it can realistically become is a secondary source, meaningful inside China, useful as a head for buyers desperate enough to overlook the premium. That brings in the geopolitical layer. A bipartisan group of US lawmakers, including Schumer and Banks, sent a letter to Apple warning Cook against purchasing from CXMT or WIME-CTC.
Both companies carry military entity designations. Apple, for its part, described current memory pricing as a once-in-a-century supply shock. The tension is clear. US policy is trying to close off Chinese suppliers precisely when the shortage makes them attractive. Apple's procurement strategy is now a test case for how far that policy pressure actually holds.
Laid on top of the IPO story is a separate development that's harder to evaluate. An unnamed Chinese firm claimed it can manufacture immersion DUV lithography machines. If true, that's a structural shift. ASML's export control monopoly depends on China not being able to build its own tooling. The signal here is real, even if the claim isn't independently verified. Mark Seyfalscher is a man of methods pro.
A three- to five-year development cycle to anything commercially competitive is the realistic floor. ASML's position outside China remains structurally secure for now, but the long-term competitive pressure is real and mounting. Here's the thing. Intel reported second-quarter results that beat revenue estimates, but the Foundry unit posted losses of $2.09 billion.
That number matters because it quantifies the restructuring cost that Intel's ambitious packaging roadmap has to justify. The EMIB-T substrate technology they're betting on for 2027 production targets costs 50% less than TSMC's co-OS. The substrate yield is currently at 50%, approaching 90% on the packaging side. Those numbers need to move significantly before customer adoption becomes real. The revenue beat is fine. The foundry execution gap is the story. At the Flash Memory Summit this week, the industry's focus landed clearly on memory tiering. Micron presented CXL-attached DRAM showing 5-10 times AI inference D-DUP compared to MVME in production tests. Keoxia dedued its CM-10 SSD combining 332-layer NAND with cold plate liquid cooling. Samsung and Micron both reached mass production on PACE 6 SSDs. The direction is unmistakable. The HBM-DRAM-NVME-CXL stack is being actively redesigned for AI workloads, and Warnkey Valley Cache is the battleground where CXL's economics will be proven or disproven.
The watchpoints from here are specific. First, whether Apple's procurement team finds a path around the CXMT and YMTC restrictions or holds the line under policy pressure. Second, whether an independent verification of China's DUV claim emerges. Third, whether Intel's EMIBT substrate yield progresses enough by mid-2027 to anchor a real customer pipeline. The CXMT surge wasn't irrational. A worsening shortage, a massive valuation, and a geopolitical flashpoint arriving together will move markets.
The signal embedded in all of it is that memory supply is now a strategic variable, not just a commodity input. That's a frame worth keeping. Thanks for listening. This podcast was built using AI technology, a YesWe production.