A 35-year-old memory-chip maker just did something no semiconductor company has done in mainland China’s stock market history: ChangXin Memory Technologies (CXMT) overtook Tencent on Thursday to become the country’s most valuable publicly traded company, with a market capitalization near $524 billion versus Tencent’s roughly $510 billion. The flip happened even as CXMT shares slipped 1.2% on the day — Tencent fell further after disclosing a 176% surge in AI-related capital spending that pushed its free cash flow negative. It’s a striking illustration of how AI investment dollars are increasingly rewarding the companies supplying the hardware, not just the ones buying it.
The numbers behind CXMT’s rise are eye-popping. The company priced its Shanghai IPO on July 27, raising $8.6 billion after its retail tranche was oversubscribed by 212 times, then surged 466% on its first day of trading. CXMT held 7.67% of the global DRAM market in 2025 and has rapidly shifted its business mix toward data-center customers — server products grew from just 8.4% of revenue in 2024 to 26.5% in 2025. The company swung to a 35.43 billion yuan (about $5.2 billion) operating profit in the first quarter, reversing a loss from a year earlier, as memory prices climbed through an ongoing global DRAM shortage. Notably, Tencent itself is a CXMT customer, having signed a $3 billion server DRAM deal in June; CXMT also inked a five-year, $7 billion agreement with ByteDance in July.
U.S. retail investors can’t buy CXMT shares directly since it trades only in Shanghai, but the story matters beyond China. It’s the clearest signal yet that the ongoing memory shortage is a durable trend, not a blip — good news for U.S.-listed memory players like Micron, whose $1 trillion market cap still dwarfs CXMT’s, and Western Digital. Investors positioned in the memory-chip supply chain are benefiting from the same AI capex wave that’s squeezing the margins of the hyperscalers doing the spending. Watch DRAM and NAND pricing trends closely — as long as they keep climbing, the suppliers, not just the AI model builders, look like the safer place to park AI-adjacent capital.
