DRAM-atic

DRAM has long been one of the most cyclical parts of the semiconductor industry. Unlike processors, where performance and design can create meaningful differentiation, memory is largely a commodity product. When demand continuously exceeds supply, prices soar; and vice versa.

Despite its commodity-like nature, the industry is protected by enormous barriers to entry. DRAM manufacturing requires specialised production processes, decades of expertise, and dedicated fabrication plants. Leading “logic” foundries such as TSMC do not manufacture DRAM. The economics of building a new DRAM business from scratch make it extremely difficult for new competitors to emerge.

As a result, the market is dominated by three companies: Samsung Electronics, SK Hynix and Micron Technology.

The industry’s latest upswing is being driven by artificial intelligence and, more specifically, a specialised form of memory known as High-Bandwidth Memory (HBM).

HBM stacks multiple memory chips on top of one another and connects them using thousands of microscopic vertical pathways known as Through-Silicon Vias (TSVs). This design dramatically increases memory bandwidth while improving power efficiency, allowing AI systems to process massive datasets more effectively. This is particularly important for large language models, where enormous amounts of data need to move continuously between memory and processors.

Among memory manufacturers, SK Hynix has emerged as the current leader. The company invested aggressively in the technology early and became Nvidia’s primary HBM supplier during the AI infrastructure boom. Its close relationship with Nvidia has helped establish a leadership position in one of the fastest-growing segments of the semiconductor industry. However, Micron has also recently secured positions in Nvidia’s latest AI platforms.