While China develops workarounds to US sanctions of high-end extreme ultraviolet (EUV) machines, the shortage in the memory industry due to high demand from AI chips has created an opportunity for Chinese firms to gain market share, suggests a report from the Korean media. The aggressive approach, led by the Chinese government, has narrowed down the gap between China and Korea for HBM memory to three years, says the report. At the center of the rapid gain is China's CXMT, which also plans to issue its shares through an IPO in order to raise capital.
China's CXMT Expected To Hit 300,000 12-Inch Monthly Wafer HBM Mass Production Capacity By 2026 End, Says Report
The gist of today's report surrounds HBM3 memory, which is the third generation of the advanced memory chips used in AI GPUs. As per the Seoul Economic Daily, Chinese firms have reached technology parity with Korean memory manufacturers Samsung and SK hynix.
HBM3 memory is used by NVIDIA's H100 AI GPUs, which were among the first that were restricted for sale by the US in China. In order to make its products compliant with the US rules, NVIDIA created the China-specific H20 GPU variant, which carries 96 GB of HBM3 memory compared to the H100's 80GB.
Chinese Government Eager To Catch Up To Global Standards Of Memory Production, Says Report
According to industry sources quoted by the Daily, China is three generations behind Samsung and SK hynix when it comes to memory production. Specifically, Chinese memory manufacturer CXMT has now achieved parity when it comes to HBM3 manufacturing. The sources add that while yield continues to be a constraint for the Chinese company, in terms of technology, CXMT is capable of producing the chips.
The parity has come as the Chinese government aggressively pushes CXMT to expand its production capacity. According to the sources, the memory manufacturer is expected to be capable of producing around 300,000 12-inch wafer units by 2026 end.
However, Samsung and SK hynix are several generations ahead of CXMT. The latest NVIDIA chips use HBM3e memory, and by the end of this year, AI GPU manufacturers are expected to enter contracts for HBM4 memory chips.
HBM4 is a major generational leap that nearly doubles the amount of data that the chips can carry over HBM3. Additionally, manufacturing AI GPUs requires complex processes such as advanced packaging available only at Taiwan's TSMC. To cash in on the memory boom, CXMT has also received approval for an IPO to raise more than $4 billion in capital.
About the author: Ramish is a seasoned technology writer and editor with more than a decade of experience. He specializes in semiconductor fabrication and market analysis. With a background in finance and supply chain management - via his bachelors in Finance and a micromasters in supply chain management from MIT - Ramish combines financial rigor with deep industry insight to deliver accurate and authoritative coverage.
Follow Wccftech on Google to get more of our news coverage in your feeds.
