Wall Street Loved the Memory Shortage. China's Latest Breakthrough Could Change That
Memory chipmakers have enjoyed being indispensable to the AI boom, but their comfortable position is getting harder to defend. Chinese chipmaker CXMT has unveiled a potentially cheaper design that could challenge established suppliers. Meanwhile, Intel and Broadcom executives are warning that the next AI bottleneck may have nothing to do with chips. Investors weren't thrilled, sending memory stocks lower on Oct. 12.
China has another idea
ChangXin Memory Technologies (CXMT) plans to ship a next-generation DDR5 server memory chip using a new 4F² architecture by the end of 2026, President Cao Kanyu announced at an industry conference in Shanghai.
The technology replaces traditional planar transistors with vertical ones, reducing the space occupied by each memory cell by 33.3%. That could increase memory density by 50% without requiring the most advanced lithography equipment.
Samsung is reportedly targeting 2028 for a similar approach, giving CXMT an ambitious two-year head start if it can deliver on schedule.
Why Micron took a hit
Samsung Electronics, SK Hynix, and Micron have long led advanced DRAM development, benefiting from the difficulty of producing memory that meets demanding server requirements.
CXMT wants a bigger piece of that business. The Chinese company held 7% of the global DRAM market in the second quarter, but its latest announcement raises questions about how much competition established suppliers could face.
Memory stocks fell on Oct. 12 as investors weighed the prospect of cheaper Chinese alternatives. Still, developing a new design and manufacturing it reliably at scale are very different challenges.
CXMT must demonstrate consistent production yields and pass strict server qualification tests before its chips can compete meaningfully. Micron's strong September earnings haven't changed, although investors are already looking further ahead.
The money problem
Broadcom CEO Hock Tan thinks the AI industry's next major constraint could be capital rather than computing power. Speaking at the DBS Asia Future Forward Conference in Singapore, Tan argued that hardware shortages eventually resolve themselves.
Financing is another matter. AI companies are spending enormous amounts building data centers, and higher borrowing costs could make those investments harder to sustain.
Intel CEO Lip-Bu Tan voiced similar concerns, pointing to the cost of capital and whether companies will eventually earn enough from AI to justify their spending.
The concern is straightforward. Building more computing capacity won't help much if customers struggle to finance the next expansion.
What investors should watch
Memory manufacturers face a potential pricing problem if CXMT succeeds in bringing more competitive chips to market. For the broader AI industry, the bigger question is whether customers can keep spending at the pace suppliers have come to expect.
Nvidia, Advanced Micro Devices, and Marvell Technology also came under pressure on Oct. 12, showing how quickly concerns about one corner of the chip market can spread.
CXMT still needs to prove its design can be manufactured at scale. Meanwhile, AI developers must demonstrate that their growing investments can produce enough revenue to support further expansion.
The irony is that chipmakers have spent years trying to solve AI's hardware shortages, only to face the possibility that customers may eventually run short of money instead.
