An AMD executive recently stated that the high prices for consumer DDR5 memory could persist until 2028, a timeline that reflects the profound impact of the AI boom on the semiconductor market.
This situation stems from a clear causal chain. First, the global AI development race has prompted hyperscalers—massive tech companies like Google, Microsoft, and Amazon—to invest hundreds of billions of dollars into building out their AI infrastructure. This investment has created unprecedented demand for high-performance memory, which is essential for training and running complex AI models.
Second, this surge in demand is heavily concentrated on specific types of memory: HBM (High Bandwidth Memory) and high-density server-grade DDR5. These components offer the speed and capacity needed for AI servers. In response, major memory manufacturers like SK hynix, Samsung, and Micron have shifted their production priorities to cater to this lucrative, high-demand market. They are reallocating their advanced manufacturing capacity, which was previously used for a mix of products, to focus on these high-margin server components.
Third, this strategic shift has created a supply squeeze for the consumer DDR5 memory used in everyday PCs and laptops. With manufacturers prioritizing server and HBM orders, often secured through Long-Term Agreements (LTAs), less production capacity is available for the consumer market. The result is a structural shortage that drives up prices for PC builders and consumers.
While companies are investing heavily in new factories and equipment—for instance, SK hynix's multi-billion dollar order for advanced EUV machines—there's a significant time lag. It takes years to build a new semiconductor fab, install the equipment, and ramp up production to a meaningful volume. Therefore, the new supply needed to balance the market is not expected to fully materialize until 2027 or 2028, aligning with the AMD executive's forecast.
- Glossary:
- Hyperscalers: A term for very large-scale cloud computing and data center operators, such as Google, Amazon (AWS), Microsoft (Azure), and Meta.
- HBM (High Bandwidth Memory): A high-performance type of memory with a very wide data bus, used in high-end GPUs and AI accelerators for faster data access.
- LTA (Long-Term Agreement): A contract between a supplier and a buyer for the long-term supply of goods at predetermined prices or conditions, often used to secure critical components.