Samsung has warned that the global memory shortage is expected to worsen in 2027, with demand continuing to outpace supply as artificial intelligence infrastructure expands at a rapid pace.
Speaking during the company’s second-quarter earnings call, Samsung said the rise of agentic AI is driving a sharp increase in token consumption, leading to unprecedented demand for memory used in both AI servers and general-purpose computing servers.
According to the company, AI developers that have struggled to secure cloud capacity from hyperscalers are increasingly turning to neocloud providers. That shift has resulted in server manufacturers purchasing memory in larger volumes, while many AI companies continue to face infrastructure constraints due to limited memory supply.
Samsung also revealed that several frontier AI developers have begun sharing medium- and long-term demand forecasts directly with the company and are requesting long-term supply agreements to secure future memory allocations.
The company believes supply will remain constrained despite higher industry investment. Samsung noted that it takes more than three and a half years for a new semiconductor fabrication plant to progress from construction to wafer production, making it unlikely that meaningful new capacity will come online before 2028.
Based on current customer forecasts, Samsung expects a significant amount of unmet demand to carry over into next year, placing additional pressure on supply.
“We expect the memory shortage in 2027 to be even more severe than it is this year, with tight supply conditions likely to persist into 2028,” the company said.
Looking further ahead, Samsung said it expects major AI customers to continue seeking multi-year supply agreements as they build long-term AI infrastructure. The company believes these agreements will help reduce the historical volatility of the memory market while allowing Samsung to expand production capacity more flexibly in line with long-term demand.
