Memory crisis in DRAM, SK Hynix expects continuing shortage until 2028

Internal documents from the South Korean memory manufacturer SK Hynix indicate that the supply situation for DRAM will remain tight until at least 2028. This primarily affects classic memory products for end users, including DDR4 and DDR5 for desktop and notebook systems as well as GDDR6 and GDDR7 for graphics cards. The analysis suggests that this is not a short-term market fluctuation, but a structural shift within the global semiconductor industry in which production priorities are changing permanently.

According to the information available, SK Hynix is increasingly gearing its production capacities towards the high-performance data center sector. The focus is on HBM memory, which is required for AI accelerators and server solutions and achieves significantly higher margins than traditional consumer memory. A return to previous production volumes for PC and graphics memory is not envisaged in the internal forecasts. I cannot verify whether the circulating documents are completely authentic, as there is no official confirmation from the company to date.

The shift to HBM has a direct impact on prices and availability in the end customer market. HBM memory is significantly more space-intensive to manufacture than conventional DRAM, as the chips are larger and require additional logic layers. This means that fewer memory bits can be produced per wafer, while at the same time the yield decreases due to more complex production processes. Even if wafer capacity remains stable or increases, the global supply of DDR and GDDR memory is therefore reduced disproportionately. For end users, this means a high probability of permanently higher prices and limited availability over several years.

SK Hynix is investing heavily in new production facilities and machinery, including several billion US dollars in additional factories and production technology. However, according to information currently available, these investments are being used almost exclusively to expand HBM production. This is not expected to provide any noticeable relief for the traditional PC and graphics memory market in the short term. The strategy is primarily aimed at meeting demand from major customers in the AI and server sectors, including manufacturers of accelerator cards. There have been no concrete commitments to increase consumer production to date.

Additional uncertainty is also evident along the retail supply chains. Occasionally, product listings have appeared with delivery dates very far in the future, in some cases as far into the year 2027. I cannot verify whether this information is based on real plans or technical placeholders. Nevertheless, market observers see such signals as an indication of a lack of planning certainty among dealers and system integrators, who are only able to calculate their procurement to a limited extent.

Overall, the available information paints a picture of a structurally changed memory market in which traditional PC and graphics memory are permanently lagging behind the requirements of the AI sector. If the strategy of SK Hynix and comparable manufacturers is confirmed, a sustained easing of prices and availability cannot be expected until at least 2028. For end users and system manufacturers, this indicates a long-term phase of limited planning security in which memory will remain a cost-relevant bottleneck.

Conclusion

The available information points to a long-term structural shift in the memory market in favor of AI and server applications. Classic DRAM for PCs and graphics cards will therefore remain a scarce commodity with corresponding price pressure until at least 2028. Even high investments by manufacturers will not necessarily lead to relief for the consumer market, as new capacities will mainly flow into high-margin specialty products. For end users and system manufacturers, this means continued limited planning security and permanently increased costs for memory upgrades.