Prediction: This Will Be the Next AI Stock to Split (Hint: It's Not Micron or Sandisk)

It seems similar everyone wants to speech astir Nvidia when the speech turns to artificial quality (AI) infrastructure. Fair enough. But there's a quieter, little glamorous portion of the AI build-out that has been generating immoderate of the astir staggering returns of 2026, and it hardly gets a notation successful the headlines: memory.

When an AI exemplary runs, it doesn't conscionable request processing power. It requires a ample magnitude of moving representation to clasp information portion computations are performed. The chips doing that occupation are called precocious bandwidth memory, oregon HBM, and the companies that marque it are drowning successful request that they can't support up with.

Missed Nvidia successful 2009? This Rare Signal Is Flashing Again. In 2009, a "Double Down" awesome flashed for a little-known chipmaker called Nvidia. For the archetypal clip successful years, that aforesaid "Total Conviction" awesome is flashing for a institution 1/100th the size of Nvidia. Continue »

That request imbalance is precisely what the Roundhill Memory ETF (NYSEMKT: DRAM) is built around, and if the exchange-traded money (ETF) keeps climbing the mode it has, a banal divided whitethorn not beryllium acold off.

A record-breaking motorboat

The Roundhill Memory ETF lone started trading connected April 2. In little than 45 days, it gathered astir $10 cardinal successful nett assets, a grounds gait for ETF plus growth. The terms went from astir $26 astatine motorboat to much than $81 astatine its peak, a astir 78% summation successful nether 2 months. As of precocious June, it has pulled backmost adjacent the $70s, but the trajectory since inception has been hard to ignore.

When an ETF's stock terms runs this acold this fast, money managers often divided the shares to bring the terms backmost into a scope that feels much accessible to mundane investors. Roundhill has done this earlier with different thematic funds. Given the DRAM ETF's trajectory, a divided is not conscionable plausible; it's likely. I deliberation it's a tenable prediction for aboriginal this year.

What's really wrong

The money holds a concentrated radical of the world's starring representation spot producers. SK Hynix and Samsung Electronics each relationship for astir 24.5% of the portfolio, with Micron Technology astatine astir 26% (including 10% successful leveraged shares) and Sandisk rounding things retired astatine 4.8%. That's astir 80% of the full money sitting successful 4 companies.

That attraction is worthy knowing earlier jumping in. This is not a diversified tech fund. It's a nonstop stake connected the representation proviso chain, and if that proviso concatenation runs into trouble, there's nary cushion from different sectors.

That said, the underlying companies person been performing successful a mode that's hard to reason with. SK Hynix banal has climbed astir 300% this year, portion Samsung is up astir 195%. The crushed is straightforward: Demand for HBM chips utilized successful AI accelerators is outpacing supply, and prices are rising arsenic a result. Every clip a hyperscaler builds retired different clump of AI servers, representation companies get a portion of that order.