The global memory chip market is facing tremendous headwinds. The key driver of this increase is the skyrocketing demand for DRAM in AI-related applications. Samsung Electronics is taking action on compelling needs. They intend to ramp up production later this decade at a new plant in Pyeongtaek, South Korea. This move is an encouraging response to high prices and supply chain disruptions. These issues have been compounded by the associated environmental impacts from the rapid establishment of AI data centers. The company’s advancements in high-bandwidth memory (HBM) technology, particularly the recently released B300, underscore the urgency to innovate and expand capacity to meet market demands.
The B300 model uses a total of eight HBM chips, where each chip is actually a stack of 12 DRAM dies. Yet another indication of this trend is where DRAM production is becoming even more concentrated to be closer to the needs of AI workloads. The amount of DRAM needed for GPUs and AI accelerators in data centers has spurred an insatiable demand. This sudden and extreme demand is yanking supply from other industries and pushing prices even further up.
Perhaps the most fruitful avenue of all innovation from Samsung has been successful showing of 16-high stack using hybrid bonding technology in 2024. According to the company’s own research, hitting a stack of 20 dies will be possible in the near future. HBM4 specification allows for up to 32 stacked DRAM dies. Compared to today’s chips which only use 12 dies, there’s obviously still room for incredible gains in technology.
The Complex Dynamics of DRAM Production
Beyond just this impact on DRAM production, the world of DRAM is a complicated one. The first has to do with how DRAM towers are physically designed. At 750 micrometers thick, these machineries of urbanism look more like a brutalist office block than their taller tower brethren. They are purposely made to be stacked on top of a base die that manages communication between memory dies and processors. This architecture is a key enabler for specifically addressing the efficient operation of AI accelerators. They mutually rely on up to 2,048 fragile micrometer-scale interconnections to exchange signals with the underlying DRAM.
Addressing supply issues in the DRAM market can be approached in two primary ways: through innovation or by constructing new fabrication plants, commonly known as fabs. It takes an act of Congress to create a fab—literally, since the process to build a new fab takes more than 18 months to cycle through. This timeline often results in new capacity landing years after the first wave of demand has subsided. In turn, the market becomes inundated and prices plummet, sometimes by drastic margins.
The fiscal implications of building these fabs are immense, with the price tag frequently over $15 billion. Consequently, businesses are discouraged from investing in growth except in the midst of booming economic times. Many firms are reluctant to expand production capacity because of the state of the economy. This timidity arrives on the heels of their recently rebounding from a late 2023 slump.
The Unprecedented Demand for AI-Driven DRAM
In fact, globally there are close to 2,000 new data centers either in planning or under construction right now. This additional growth is indicative of a never-ending, bottomless pit demand for computing power led almost exclusively by the AI workloads. Recent numbers paint an astonishing picture of Nvidia’s dive into data center dollars. It shot up from a little over $1 billion in late 2019 to a staggering $51 billion by October of 2025.
Analysts note their concern over the DRAM market’s supply chain sustainability and pricing structure. It’s a legitimate concern and one that comes from the rapid increase in demand. According to Mina Kim, an industry expert, “In general, economists find that prices come down much more slowly and reluctantly than they go up. And DRAM today should be no exception to this general observation, particularly with the continuing, almost infinite appetite for compute.”
Manufacturers are focusing their production capabilities to directly address the increased demand. They need to adopt technological innovation and find a prudent middle ground between continuously expanding new capacity and preparing it for market usage.
The Future Outlook for Memory Chip Supply
Experts are cautiously optimistic about technology and production pipeline. They don’t expect us to get any relief from the ongoing memory chip shortage before 2028. As Intel CEO Lip-Bu Tan put it very simply, “There’s no relief until 2028.” This sobering forecast serves as a reminder of the challenge still facing manufacturers and consumers as they all grapple with persistent supply shortages.
Shawn DuBravac, another industry analyst, noted that future relief will likely arise from a combination of factors: “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.” Together, this multifactored plan should alleviate at least some of the strain that our industry is feeling today.
Companies such as Samsung are making strong strategic investments and innovative leaps to position themselves toward future demand. At the same time, stakeholders are closely watching market conditions and new innovations that can change the landscape of memory chip manufacturing.