Memory chipmakers face boom-bust volatility as AI-driven demand cycle collides with cyclical oversupply risk in HBM and DRAM markets.

It's a good time to be in the memory business. As the AI datacenter sector booms, SK Hynix and Micron's revenues have tripled in the last year, while Samsung's have roughly doubled. Yet the deck is stacked for reversal—such is the memory business historically.

Today, sky-high demand for high-bandwidth memory (HBM), DDR5, and NAND flash memory needed for GPU servers has consumed all remaining capacity, creating shortages that have driven up prices across consumer electronics and AI infrastructure alike. Budget smartphones are now difficult to purchase at affordable prices.

The big three memory vendors are investing hundreds of billions of dollars to bring new fabrication capacity online. In June, South Korean President Lee Jae Myung announced a $576 billion investment led by SK Hynix and Samsung to bolster chip production and shore up AI supply chains. Micron said Thursday it would invest up to $3 billion to strengthen the US semiconductor supply chain, while also working to boost production at sites in Singapore, Taiwan, and Japan.

Semiconductor manufacturing is among the most complex and resource-intensive industries in the world. Building a new DRAM or NAND flash wafer fab requires securing financing, selecting a location, winning permits, and deploying tens of millions of dollars in support facilities—power conditioning, air handling, ultra-pure water filtration systems. After clean rooms are completed, hundreds of millions more in specialized lithography, wafer transport, and test equipment must be installed and validated. Once powered on, it can take months to dial in settings and bring yields to acceptable levels. This process often takes years even without delays.

While a handful of new memory fabs are already under way, anything SK, Samsung, or Micron starts today will take at least three years to bring online, and even longer to ramp production. A recent IDC report warns that relief from the memory shortage may not arrive until at least 2028.

That's excellent news for memory makers, whose revenues will remain elevated. For AI startups and model developers, however, it's a significant problem. They will continue paying higher infrastructure prices, and sky-high memory costs are not conducive to finding margins in the cost per token. OpenAI and other ventures have spent the last four years and hundreds of billions in venture capital developing increasingly capable models, agents, and tools. It's no longer a question of whether the technology works, but whether the benefits justify continued investment at current or higher levels. Sooner or later, these startups must turn a profit—and elevated memory prices certainly aren't helping.

The central question is whether memory vendors can bring new capacity online before the major AI houses exhaust their VC-subsidized runway and the music stops.

Historically, memory is a commodity with wild swings in pricing characterized by boom-and-bust cycles. Memory vendors therefore rely on boom cycles to finance new fabs, knowing full well that once capacity comes online, the additional supply could crater prices. As reported last year, the AI boom has changed this dynamic dramatically. Rather than the expected price declines across 2025 and 2026, prices have risen steadily as AI infrastructure consumes every bit of DRAM and NAND available.

If anticipated demand for AI falls short, however, everyone loses. Memory vendors will find themselves at the bottom of a bust cycle of historic proportions.

On a bright note, sky-high memory prices will no longer be the reason you can't afford a new laptop or smartphone.