The current scramble for memory chips—the essential components that allow computers to process data quickly—might be nearing its end much sooner than expected. While most industry analysts point to a prolonged shortage lasting until 2028 or even 2030, a new forecast suggests a significant shift in supply could arrive as early as late 2027.

A potential shift in the RAM landscape

Kye-hyun Kyung, the former head of Samsung's semiconductor business, recently shared a more optimistic timeline during a keynote at the National Academy of Engineering of Korea in Seoul. He noted that aggressive production expansions by Chinese companies are set to change the market dynamics. According to reports from Wedbush, companies like CXMT are targeting significant capacity increases, including a push for HBM3 (a high-speed memory type used specifically for AI) production by 2026–2027.

Kyung suggested that a surge in memory supply could begin in the second half of 2027 or the first half of 2028. This would provide a much-needed relief to a market currently squeezed by the sudden demands of the AI boom. As we previously explored in our earlier piece on rising memory costs, the intense hunger for high-bandwidth memory has fundamentally reshaped how manufacturers allocate their output.

Before assessing the validity of this timeline, it is important to look at the specific expansion targets being set by Chinese manufacturers. These figures represent the intended scale of upcoming operations rather than current output.

The scale of Chinese expansion

Chinese manufacturers are moving to bridge the gap in domestic production. YMTC, a major player in the NAND market (the type of memory used for long-term storage like SSDs), is planning to more than double its capacity. Sources indicate they aim for approximately 500,000 wafers per month by late 2026 or 2027, utilizing three new fabrication plants in Wuhan. For context, YMTC's baseline capacity was roughly 200,000 wafers per month prior to these planned expansions.

Similarly, CXMT is working toward a capacity of roughly 300,000 wafers per month for DRAM (the short-term memory used by CPUs) by the 2026–2027 window. While these targets are ambitious, it is worth noting that actual production "ramp rates"—the speed at which a factory reaches full capacity—often lag behind official announcements.

Countervailing forces and risks

Despite this massive push for capacity, several hurdles remain that could prevent a total market correction. One primary obstacle is the ongoing restriction of high-end technology. U.S. export controls continue to limit the access Chinese manufacturers have to the advanced tools required to build the most sophisticated chips.

To bypass these limits, Chinese firms are increasingly relying on "localization"—building their own domestic supply chains for manufacturing equipment. Companies like Naura and AMEC are working to provide the tools necessary to scale production despite international restrictions. This shift is a strategic attempt to decouple their growth from Western technology.

There is also a question regarding the economic sustainability of the current AI trajectory. Kyung observed that the return on investment for "Big Tech" could potentially decrease relative to the massive amount of capital being spent on AI infrastructure. If the profitability of AI services does not match the investment, the demand for high-end memory might soften, further accelerating the supply surge.

While the current consensus from major players like Samsung suggests a shortage lasting until 2030, Kyung's perspective offers a different possibility. Will the rapid scaling of Chinese production be enough to break the current supply constraints, or will regulatory hurdles keep the market tight for years to come?