Worldwide demand for Dynamic Random-Access Memory (DRAM) has exploded. This increase is mostly driven by the skyrocketing use of artificial intelligence (AI) programs. The rapid increase in demand has drawn memory supplies from other industries as well. Consequently, costs have rapidly escalated nationwide. Even innovators like Samsung are retrenching production in reaction to shifting market dynamics. At the same time, exciting new breakthroughs in DRAM technology promise to transform the landscape of memory supply.
Data centers are ground zero for this exploding demand. As corporations have rushed to build more powerful AI solutions, demand for advanced memory technologies has only grown. The current shortage has revealed a classic boom and bust condition endemic to the DRAM industry. It happens to be occurring alongside an unprecedented expansion of AI hardware infrastructure.
NVIDIA emerges as the crucial villain. The data center business revenue jumped up from a little over a billion dollars in Q4 2019. By October 2025, it’s on pace to exceed about $51 billion! This expansion reflects our greater dependence on high-performance memory solutions. In fact, there are almost 2,000 new data centers planned or in construction globally.
Production Cuts and Strategic Moves
Looking at the rapidly inflating demand and unpredictable investment climate, Samsung introduced price control to sharply stabilize prices. The company has made headlines this past week with its announcement of plans to cut memory chip production by 50 percent. This new rule aims to stop prices from dropping below the cost of production. It also showcases the extreme competitive pressures of the DRAM market.
Samsung’s determination to demonstrate both innovation and leadership is clear as the company focuses on bringing new technology to market first over the next few years. To increase its manufacturing capacity, the company is now working on HBM4 (High Bandwidth Memory). This cutting-edge memory is capable of stacking up to 16 DRAM dies on top of each other. This development follows Samsung’s milestone in 2024 for producing a 16-high stack using hybrid bonding technology. In addition, Samsung executives indicated that a 20-die configuration, more than doubling the memory bandwidth, might be possible.
“Relief will come from a combination of incremental capacity expansions by existing DRAM leaders, yield improvements in advanced packaging, and a broader diversification of supply chains,” – Shawn DuBravac
Analysts are warning that buyers shouldn’t expect immediate relief from the supply crunch. Building greenfield fab facilities now requires more than 18 months and requires extreme capital intensity of over $15 billion per facility. Consequently, an estimated timeline for when supply chain issues will be remediated remains uncertain at best.
Mina Kim, an expert in semiconductor economics, pointed out two primary strategies for alleviating DRAM supply challenges: innovation through advanced technology and building additional fabs.
“There are two ways to address supply issues with DRAM: with innovation or with building more fabs,” – Mina Kim
The Complex Dynamics of Memory Technology
The complexities of today’s memory technology only add to the tumultuous supply landscape. High Bandwidth Memory (HBM) modules are coming to be considered essential in maximizing performance with AI workloads. These memory modules usually plug in on both sides of processors. This architecture permits great compactness, packing GPUs and memory into one super-connected chip. The picture gets complicated very quickly when DRAM is only a millimeter away from AI accelerators. They interconnect through literally thousands of micrometer-scale links (2,048 maximum per core).
China’s recently released B300 server reflects some dramatic improvements in technology. It leverages eight HBM chips (12 DRAM dies each) to meet the skyrocketing needs of AI workloads that are power-hungry. Such breakthroughs will be needed as the industry continues to struggle to meet an overwhelming increase in demand with limited supply.
Even with these technological advances, experts are doubtful about how quickly and far prices return to normal.
“In general, economists find that prices come down much more slowly and reluctantly than they go up. DRAM today is unlikely to be an exception to this general observation, especially given the insatiable demand for compute,” – Mina Kim
Looking Ahead: The Future of DRAM Supply
Samsung really does have incredible stuff planned for the future! They’ll pump up fabrication at a new plant in Pyeongtaek, South Korea, which is scheduled to come online in 2028. This expansion is essential not just for addressing long-term supply challenges, but for keeping pace with the current dynamic evolution of AI technologies. Even industry leaders concede they’re not expecting major relief from the grueling present until then.
Intel’s CEO, Lip-Bu Tan, succinctly characterized the situation facing the semiconductor industry:
“There’s no relief until 2028.”
As the demand for DRAM continues to rise amid rapid developments in AI technologies, stakeholders across the semiconductor landscape are left navigating a complex web of challenges. Innovation and capacity expansion will be key ingredients on how the industry will answer the challenge. They will shape how well and how swiftly we rise to meet these extraordinary challenges.