Nvidia (NASDAQ: NVDA) CEO Jensen Huang has made a pointed observation about the state of semiconductor supply chains that every investor should take seriously.

During a recent trip to South Korea, Huang stated, “The whole industry supply chain — everything from wafers to packaging to silicon photonics…everything’s in short supply because the demand is so high. It is going to persist for several years.”

That assessment reflects a structural shift in how memory is valued across the global technology landscape, moving far beyond its traditional role as a commodity input.

Artificial intelligence has transformed memory into one of the most critical and constrained resources within hyperscale chip stacks, driving what analysts are calling a full-scale memory supercycle.

Training generative AI models and running inference deployments require enormous bandwidth between processors and memory, with high-bandwidth memory, or HBM, emerging as the key enabling technology.

HBM is an advanced form of dynamic random-access memory layered directly with graphics processing units, delivering the extreme data transfer speeds that modern AI workloads demand.

Hyperscalers including Alphabet, Amazon, Microsoft, and Meta Platforms are spending hundreds of billions of dollars annually to build next-generation data centers, with a growing share of those budgets allocated to memory and storage.

Unlike previous memory cycles, where demand was tied to price-sensitive consumer devices and PC refresh cycles, today’s AI-driven demand is treated by hyperscalers as a critical infrastructure input rather than a cost to minimize.

On the supply side, HBM production is far more complex than standard DRAM manufacturing and requires multiyear fabrication investments, meaning supply growth is unlikely to keep pace with compounding AI demand anytime soon.

For investors seeking broad exposure to this theme without the risk of picking individual winners, the Roundhill Memory ETF (NYSEARCA: DRAM) offers a pure-play solution across the global memory industry.

The fund holds a diversified portfolio of memory and storage leaders, including Micron Technology, Samsung Electronics, SK Hynix, Kioxia, Sandisk, Seagate, and Western Digital, among other specialists.

Its geographic reach spans the United States, South Korea, Japan, and Taiwan, providing meaningful diversification across the world’s most important semiconductor manufacturing hubs.

With an expense ratio of 0.65%, the Roundhill Memory ETF delivers targeted exposure to the AI memory trade at a relatively low cost for investors.

If the memory shortage persists for several years as Huang suggests, companies across the DRAM supply chain stand to benefit significantly from sustained pricing power and rising capital investment.

The convergence of structural supply constraints and exponentially growing AI demand makes the memory sector one of the most compelling infrastructure plays in the current market cycle.