A new analysis from Citrini predicts a substantial shortfall in the global DRAM market by 2030, with demand expected to exceed supply by a significant margin. Even with increased production from Chinese manufacturers like CXMT, the gap is projected to be approximately 28.7 exabytes (EB) out of a total global demand of 157.5 EB.
This projected deficit, which represents an 18% to 25% shortfall, includes High Bandwidth Memory (HBM) but asserts that general DRAM remains the primary bottleneck. The research suggests that even current estimates for data center and AI demand might be conservative, potentially being balanced out by future demand from physical AI applications such as autonomous vehicles or humanoid robots.
To put these figures into perspective, current global DRAM capacity for 2024 is estimated at around 40 EB. The predicted unmet demand in 2030 is a significant portion of this yearly capacity. Notably, the projections show that the anticipated consumer demand for DRAM is smaller than the unfulfilled demand, indicating that personal computing needs might be secondary to the vast requirements of AI and data centers.
According to researcher Zephyr, the general DRAM market could face a 25% deficit, with an annual supply of 91 EB versus a demand of 120 EB. This imbalance is expected to keep DRAM Average Selling Prices (ASPs) inflated, likely maintaining a range of $1.50 to $2.00 per gigabit.
While the situation appears stark, potential technological advancements could mitigate the shortage. Innovations focused on AI efficiency, such as Google's TurboQuant algorithm which promises a sixfold reduction in memory overhead, might help lessen the demand. However, it remains an open question whether such efficiencies will truly reduce overall demand or simply enable further expansion of AI capabilities.