The artificial intelligence story so far is centered around immense technological progress in a short amount of time. (Three years ago, no one had heard of ChatGPT—today, few people haven’t at least tried it out).
From an economic perspective, AI is also a classic example of infrastructure bottlenecks. Out of nowhere, there’s huge demand for certain commodities and products. But they can’t be created on a whim, so many firms must wait for supply. The companies that are lucky enough to get their hands on coveted material pay staggering prices to secure supply.
The upside? There are incumbent players who benefit from these logjams—and investors can potentially profit from having exposure to them. In the early days of the current AI era, investors gravitated towards Nvidia for exactly this reason.
In the current AI landscape, the big opportunity seems to be in memory production. By now, stories of AI applications hallucinating and inventing facts are commonplace. Recently, for example, the New York Times was forced to admit that a quote it attributed to a Canadian politician never was spoken (the reporter used AI as a shortcut and got burned).
Why does AI do this? In a nutshell, first generation AI chips engage in stream-of- thought writing. They take whatever was last said and project it into the future, without historical context that checks for accuracy. The way to fix this is to add memory to the equation, so that AI has a good record of what it was asked—and how it answered—in the past. Even Nvidia needs to add memory to its GPUs to make them work efficiently.
Alas, memory production is incredibly concentrated among a few firms: Only three companies around the world are responsible for 97% of production: Samsung, SK hynix, and Micron. (These firms produce what is known as DRAM memory.)
Other than Micron, there is virtually no U.S. production of DRAM because prior to generative AI the industry became incredibly commoditized. It simply wasn’t worth it for American companies to produce these chips.
As it stands, Samsung, SK hynix, and Micron are all booked solid till 2029. And supply is very inelastic: Standing up a new factory takes at least 1-2 years, so a wave of new production is not yet on the horizon.
Investors Gravitate to the Seoul of Memory Production
Tellingly, two of the three major memory producers (Samsung and SK hynix) are based in South Korea. For most investors, getting direct access to South Korean equities is either impossible or difficult. So, to obtain exposure to these two dominant chipmakers, the workaround has been to buy ETFs linked to an index of the country’s equity market.
But that move is less than ideal. When an investor essentially owns a basket of South Korean equities, they’re not simply getting exposure to the two big memory producers. They’re also stuck with owning equities in completely unrelated industries as well. Indeed, while over 61% of the MSCI Korea Index is comprised of Information Technology stocks, there’s a raft of other sectors such as industrials and financials that account for the balance.
Introducing KMEM: The Purest Play on Memory Production
When investors sense a compelling opportunity, diluted exposure can be less than ideal. That’s the thinking behind the Kurv Memory Select ETF (KMEM). Launching on July 1 2026, KMEM seeks to obtain– via stocks, swaps and options– concentrated exposure to the companies dominating memory chip production. This includes the “Big 3” memory manufacturers (SK hynix, Micron, and Samsung), along with additional firms that play a major role in the memory ecosystem. These stocks may benefit from the existing and expected shortage of DRAM memory production. As such, KMEM offers investors an avenue for profiting from the critical layer powering AI infrastructure and next-generation computing workloads.
Once launched, KMEM will be the purest play on what may turn out to be an unforgettable opportunity.
