The artificial intelligence (AI) revolution extends far beyond which chip designer makes the best graphics processing units (GPUs). At the core of AI workloads are memory and storage -- the DRAM and high-bandwidth memory (HBM) chips that keep massive training clusters fed with data at high speeds. As data centers scale to support larger AI models, memory demand is shifting from cyclical commodity swings to a sustained structural boom.
On April 2, the Roundhill Memory ETF (DRAM +4.03%) was launched. In its first 27 trading days, the fund amassed $6.5 billion in assets under management (AUM), making it the fastest launch of any ETF in history. The ETF IPO'd at $28 a share and is already trading at just over $60.
DRAM packages the world's leading memory chipmakers into a single U.S.-listed vehicle, giving investors a streamlined way to benefit from the ongoing AI memory supercycle.
Image source: Getty Images.
Who are the key players of the AI memory supercycle?
Hyperscale AI workloads require up to 10 times more DRAM to keep GPU clusters humming without latency bottlenecks. HBM has become a must-have resource, powering everything from large language model (LLM) training to inference deployments at scale. These dynamics have shifted supply and demand across the AI infrastructure landscape, handing memory manufacturers unprecedented pricing power.
Roughly three-quarters of the Roundhill Memory ETF's portfolio is invested in three memory stocks: South Korea's SK Hynix and Samsung Electronics, alongside U.S.-based Micron Technology (MU +1.14%). Roundhill captures exposure to NAND flash leaders such as Kioxia and Sandisk (SNDK +6.55%), as well as storage specialists like Seagate Technology and Western Digital.
Why global diversification is important
One of DRAM's most interesting features is its composition. Investors gain immediate stakes in the Korean duopoly of Samsung and SK Hynix -- names that are challenging for U.S. retail (and even many institutional investors) to access, as neither offers American depositary receipts (ADRs) at this time.
It seems like AI memory demand evolves each passing quarter as new model architectures and inference workloads are released. While this ETF is less diverse than your average fund (74% of its holdings are in just three stocks), investing in the Roundhill Memory ETF provides diversified exposure in other ways, like across U.S. memory innovation (Micron), Korean manufacturing, and supporting players across NAND and storage. This level of convenience matters in thematic investing.
Moreover, diversification helps mitigate the risks of investing in a single stock. Even if one player posts monster earnings, the fund captures broader upside across the entire memory value chain.
NYSEMKT: DRAM
Key Data Points
The Roundhill Memory ETF comes with a catch
A portion of the fund's exposure -- particularly to Micron -- comes through total return swaps and leveraged derivative contracts in addition to direct share ownership. These instruments are designed to enhance returns through amplified exposure. Smart investors know that swaps and leveraged contracts can also magnify losses during corrections.
This structure is not uncommon in heavily concentrated, niche ETFs and, in my opinion, should not be a deal-breaker. All told, the Roundhill Memory ETF remains a compelling vehicle because it solves the access problem of foreign equities while also capturing one of the clearest secular movements in AI semiconductors right now.
Moreover, DRAM's record-shattering launch underscores widespread conviction that memory and storage are becoming an indispensable backbone supporting AI infrastructure build-outs. I think DRAM is a great position for investors who want targeted exposure to a booming pocket of the AI hardware stack without the risks and hassle of choosing a single stock.
