CXMT Begins HBM3E Risk Production, Marking a Major Step for China’s AI Chip Ambitions
China’s semiconductor industry has reached an important milestone as CXMT has reportedly started risk production of HBM3E memory, a key technology used in advanced AI accelerators and data center hardware. The move could significantly strengthen China’s domestic chip supply chain at a time when high-bandwidth memory has become one of the most valuable components in the global AI race.
HBM, or High Bandwidth Memory, is critical for artificial intelligence workloads because it delivers far higher data throughput than standard DRAM. Modern AI processors rely on this memory to move massive amounts of data quickly, making HBM one of the most in-demand technologies in the semiconductor market.
CXMT’s entry into HBM3E production is especially notable because China has been working to reduce dependence on foreign memory suppliers. Although the company is still behind leading South Korean memory makers, which are already preparing for newer HBM generations, producing HBM3E domestically is a major achievement for China’s chip ecosystem.
The exact specifications of CXMT’s HBM3E chips have not been officially disclosed, but HBM3E memory typically follows JEDEC standards. That means the chips are expected to use a 1,024-bit interface with speeds ranging from around 9.2 Gbps to 12.4 Gbps per pin. Many HBM3E products target roughly 9.6 Gbps per pin, which can provide about 1.2 TB/s of memory bandwidth.
Capacity will likely depend on the stack configuration. An 8-high HBM3E stack can offer around 24 GB of memory, while a 12-high stack can reach approximately 36 GB. These specifications make HBM3E well suited for AI training, inference, high-performance computing, and other bandwidth-heavy workloads.
For now, CXMT is reportedly in the risk production stage. This phase usually means the company is testing early manufacturing runs, refining production processes, and working to improve yields. In semiconductor manufacturing, risk production does not mean immediate large-scale output, but it is often the final step before volume production begins.
CXMT has gained attention for moving quickly from early production to larger-scale manufacturing. If the company follows that pattern with HBM3E, mass production could arrive sooner than many expect, assuming yield rates improve fast enough.
The company has also been expanding its manufacturing footprint aggressively. CXMT currently operates major 12-inch wafer fabs in Hefei and Beijing, with combined production reportedly reaching around 200,000 wafers per month. By the end of the year, output could climb to approximately 350,000 wafers per month.
Longer term, CXMT is aiming for much larger capacity. With additional facilities planned in Shanghai and Hefei, the company’s monthly wafer output could eventually rise to around 600,000 wafers. That scale would give China a much stronger position in both conventional DRAM and advanced memory production.
However, the breakthrough may come with a downside for consumers. HBM3E is far more complex and resource-intensive to manufacture than regular DRAM because it requires stacking multiple memory dies into a single package. As AI companies continue to demand more high-bandwidth memory, manufacturers may prioritize HBM production over standard DDR5 memory.
That could affect the consumer PC memory market. CXMT recently began supplying DDR5 chips for consumer RAM modules, raising hopes that increased competition might help reduce DDR5 prices. But if much of the company’s production capacity is redirected toward HBM3E for AI servers and data centers, the expected drop in consumer DRAM prices may be delayed.
In other words, CXMT’s HBM3E progress is a major win for China’s AI hardware ambitions, but it may not bring immediate relief to PC builders, gamers, and everyday consumers looking for cheaper RAM. If demand for AI memory remains strong, standard DRAM prices could stay elevated or even rise further over the next few years.
CXMT’s move into HBM3E shows how rapidly the memory market is changing. High Bandwidth Memory has become a strategic technology, not just a performance upgrade. As AI infrastructure expands worldwide, companies that can produce advanced HBM at scale will play an increasingly important role in the global semiconductor industry.
