Supply Tightness in the Memory Market — Explained (Part 1)

Memory Is Surging Alongside AI

Gold and silver bullion are not the only commodities reaching record highs. Memory prices are rising as well.

In recent years, demand for DRAM has surged as artificial intelligence (AI) systems expand at an unprecedented pace. Large language models (LLMs) such as ChatGPT and Claude require enormous amounts of memory to train and run, and the data centers that power these systems consume DRAM at a scale rarely seen before.

At the same time, high bandwidth memory (HBM) has been in short supply since around 2024. HBM is a specialized form of stacked DRAM used inside the GPUs that train and serve modern AI models. Companies building AI infrastructure have been purchasing these components aggressively, placing pressure on the global memory supply chain.

From Niche Constraint to Broad Impact

Until recently, memory shortages were mostly visible to niche communities. PC gaming enthusiasts were often the first to notice price spikes in graphics cards or RAM modules when supply tightened.

The current cycle is different.

The rapid expansion of AI infrastructure means that demand for memory is coming from hyperscale cloud providers, AI startups, and enterprise data centers all at once. As a result, the effects of the current RAM crunch are beginning to spread beyond hobbyists and into the broader economy. From Niche Constraint to Broad Impact

So why do AI companies need so much DRAM?

The answer begins with GPUs. Modern AI models rely on GPUs that use HBM to perform the massive matrix calculations that underlie neural networks. HBM sits directly next to the GPU and provides the extreme memory bandwidth required for these computations.

However, GPUs do not operate in isolation.

They are embedded in large server systems that manage data, coordinate workloads, and deliver AI services to users. While GPUs and HBM provide raw computational power, these surrounding systems rely heavily on conventional DRAM.

As companies build ever larger AI clusters, the amount of DRAM required to support these systems scales rapidly alongside compute.

A Foundational Shift

In an unfortunate reality for the average consumer, the current RAM crunch reflects a broader shift in the technological economy.

Just as gold and silver once served as foundational commodities for financial systems, memory is becoming a foundational commodity of the AI age.

GPUs may receive most of the attention, but they cannot operate without vast quantities of DRAM and HBM supporting them. As companies race to build larger AI models and deploy them globally, memory demand continues to rise alongside computational demand.