China's largest memory chipmaker, ChangXin Memory Technologies (CXMT), has swung to a hefty third-quarter profit, driven by a surge in demand for artificial intelligence that has lifted prices for standard memory chips. The company also signaled it is moving into the lucrative market for high-bandwidth memory (HBM), the ultra-fast chips used in AI servers.
A profit windfall from AI
CXMT reported a net profit of 54.4 billion yuan (about $7.6 billion) for the third quarter, on revenue of 118.71 billion yuan. The figures, reported by Reuters, mark a dramatic turnaround for the company, which had been loss-making in previous quarters as it invested heavily in expanding production capacity.
The profit surge is largely attributed to a shift in strategy by the industry's dominant players—SK Hynix, Samsung Electronics, and Micron Technology. These companies have been reallocating their production capacity toward HBM, the specialized memory chips that are essential for AI accelerators like Nvidia's GPUs. By focusing on HBM, they have reduced output of standard DRAM (dynamic random access memory), which is used in PCs, smartphones, and servers. This has tightened supply and pushed up prices for standard DRAM, benefiting CXMT, which primarily produces these conventional chips.
DRAM prices have been on the rise for several quarters, as the AI boom has created a surge in demand for memory across data centers and edge devices. For CXMT, which has been ramping up production to challenge the established players, the higher prices have translated into a significant boost to its bottom line.
Moving up the value chain
CXMT is now aiming to capitalize on the HBM opportunity itself. The company said it plans to start limited production of HBM3E, the latest generation of high-bandwidth memory, within two months. HBM3E offers higher bandwidth and lower power consumption compared to earlier versions, making it the preferred choice for AI training and inference workloads.
Entering the HBM market is a strategic move for CXMT, as it allows the company to tap into the most profitable segment of the memory market. HBM chips are significantly more expensive than standard DRAM, and margins are much higher. However, the technology is complex, and CXMT will face stiff competition from SK Hynix, Samsung, and Micron, which have been shipping HBM for years.
For China, CXMT's progress in HBM is also of strategic importance. The country has been seeking to reduce its reliance on foreign memory chips, especially in the context of US export controls that have restricted China's access to advanced semiconductor technology. CXMT's ability to produce HBM domestically could help China secure a more self-sufficient supply chain for AI infrastructure.
What it means for investors
For everyday investors, CXMT's results are a reminder of how deeply the AI boom is reshaping the semiconductor industry. The demand for AI has not only boosted the fortunes of chip designers like Nvidia but also the memory makers that supply the components. As AI models become more complex, the need for high-bandwidth memory is only expected to grow.
However, investors should be cautious about extrapolating CXMT's profit surge into a long-term trend. The memory chip market is notoriously cyclical, and prices can swing sharply as supply and demand shift. The current tightness in standard DRAM is partly a result of the industry's pivot to HBM, but if that capacity shift reverses or if demand for AI slows, prices could fall just as quickly.
For those invested in the broader semiconductor sector, CXMT's move into HBM is a signal that competition in the high-end memory market is heating up. That could put pressure on the margins of the established players, even as overall demand remains strong. It also highlights the growing role of Chinese companies in the global chip supply chain, despite ongoing geopolitical tensions.
As always, investors should focus on the fundamentals of the companies they own and consider the long-term drivers of demand, rather than reacting to short-term price movements. The AI-driven memory boom is a powerful tailwind, but it is not without risks.
