I run e-commerce operations in China.
That's not a disclaimer, it's context and its what I do. I'm not writing about China's semiconductor ambitions from a distance, reading think tank reports and analyst notes. I watch Chinese tech companies operate in real time. I see what they're buying, what they can't get, and how they're working around constraints that most Western investors don't fully understand.
The memory shortage is real on the ground in China. And the walls keeping Chinese memory makers from solving it are more difficult than most people realize.
There are three of them.
Wall One: They Can't Buy It
Start with the most immediate constraint. U.S. export controls block China from purchasing advanced HBM from Samsung, SK Hynix, and Micron. The Foreign Direct Product Rule extends these restrictions to South Korean firms operating in China, meaning the ban isn't just on American companies. It covers the entire Western memory supply chain.
HBM2E and abovex the memory that matters for AI, is completely off limits. HBM3, HBM3E, HBM4, the chips powering Nvidia's AI accelerators and the hyperscaler buildout. China simply cannot buy from the companies that make them. The demand exists. The willingness to fork out cash exists. The legal pathway does not.
This creates a bifurcated global memory market. Western AI builders compete for a supply pool that explicitly excludes China. Every chip that would have gone to a Chinese AI lab stays in the Western ecosystem instead, tightening supply further and keeping pricing power with Samsung, SK Hynix, and Micron.
Wall Two: They Can't Make It
This is where most analysts stop. Export controls exist, China can't buy, end of story. But the more important question is whether China can manufacture its own way out.
The answer is not yet, and the gap is wider than China's official narrative suggests.
ChangXin Memory Technologies, or CXMT, is China's most advanced DRAM manufacturer. They've produced HBM3 samples and delivered them to Huawei. That's real progress. But samples are not volume production, and volume production is not competitive yield, and competitive yield at leading-edge nodes is not something CXMT is close to achieving.
The technology gap between CXMT and the Big Three is roughly six to eight years, that’s approximately three full HBM generations. SK Hynix is ramping HBM4. CXMT is celebrating HBM3 samples. That's not a gap you close quickly regardless of how much government money you throw at it. Semiconductor manufacturing is accumulated process knowledge, engineers who've run these tools for years, yield improvements built incrementally over thousands of wafer starts, institutional expertise that can't be replicated by funding alone.
I see this in China every day. The ambition is genuine. The capability gap is equally genuine.
Wall Three: They Can't Get the Tools
The most underappreciated wall is the equipment gap.
Advanced HBM manufacturing requires extreme ultraviolet lithography machines, EUV made by ASML in the Netherlands. China cannot buy them. The Dutch government, under sustained U.S. diplomatic pressure, has blocked ASML from exporting EUV equipment to China. Without EUV, you cannot manufacture at the nodes required for competitive HBM.
But it goes deeper than ASML. The semiconductor equipment supply chain involves hundreds of specialized tools, deposition equipment, etching systems, metrology tools, chemical mechanical planarization machines, many of which are now subject to export controls from the U.S., Japan, and the Netherlands simultaneously. The three largest semiconductor equipment exporters in the world have coordinated restrictions on China due to U.S. policies.
CXMT has reportedly stockpiled enough equipment to continue current production through 2026 or 2027. After that, their ability to advance to next-generation nodes hits a hard wall. They can build more of what they already know how to build. They cannot easily develop what comes next without new equipment they cannot legally acquire.
What This Means In China’s Reality
Three walls. Can't buy it. Can't make it at scale. Can't get the tools to make it better.
Each wall alone would be a significant constraint. Together they create a protected window for Western memory makers that extends comfortably through 2027 and likely beyond. New fab capacity from Samsung, SK Hynix, and Micron takes three to five years to come online. China's domestic alternative is years behind on technology and facing an equipment ceiling.
From the ground in China, I can tell you the frustration is real. Chinese AI companies want this memory. They're paying premium prices for whatever legacy DRAM they can legally access. They're engineering around constraints in ways that are impressive but ultimately insufficient for frontier AI workloads.
The bears who say China will flood the memory market haven't looked carefully at the three walls standing between that prediction and reality.
Until those walls come down, and I'll be watching closely from inside the market, the pricing power stays exactly where it is.
The US didn't just sanction a product for China. It sanctioned an entire technological future.
I hold a long position in DRAM ETF (ticker: DRAM). Not investment advice. Do your own research.
My current positions as of May 13, 2026. Memory had profit taking yesterday, so added to my DRAM position and sold SNDK put x 1 + MU put x 1