AI ETFs: Memory & Photonics Move Into Focus
Key Takeaways:
- AI investing is broadening beyond processors to memory and photonics, which address critical data storage and connectivity bottlenecks.
- Both memory and photonics ETFs offer targeted access to global leaders, but concentrated holdings can amplify both opportunity and volatility.
- Advisors should compare ETFs carefully, since products range from pure-play equities to broader supply-chain and leveraged strategies.
The AI trade has faced renewed volatility as investors question AI spending and current valuations. Semiconductor stocks and related areas — including memory, networking, photonics, and chip equipment—have all been caught in the pullback.
From an ETF perspective, however, the theme is expanding rather than disappearing. The first phase of the AI trade largely centered on processors and broad semiconductor funds. Newer ETFs are targeting the less visible technologies needed to support those processors, particularly memory chips that store and supply data and photonics systems that help move it. These products give investors more precise exposure to the AI infrastructure buildout. However, that precision can also bring greater concentration and risk.
Why AI Needs More Memory
Memory chips help computers store and access the information needed to perform tasks. DRAM, or dynamic random-access memory, serves as short-term working memory, while NAND flash provides longer-term storage in products such as solid-state drives.
High-bandwidth memory, or HBM, has become especially important for AI. HBM is an advanced form of DRAM, designed to move large amounts of data quickly between memory and AI processors. That makes memory a critical part of the AI infrastructure buildout rather than another type of semiconductor.
Micron estimates that the addressable market for HBM could grow from approximately $35 billion in 2025 to around $100 billion by 2028, representing an annual growth rate of roughly 40%. This illustrates how quickly memory is becoming a larger component of the AI story.
The opportunity is not limited to HBM. AI servers also require conventional DRAM and substantial amounts of NAND-based storage. At the same time, memory remains a historically cyclical industry. Periods of limited supply can support higher prices and margins, while capacity, inventory issues, or weaker technology spending can reverse those conditions. Memory ETFs provide targeted exposure to a potential AI bottleneck, but because of their cyclicality, they should not be mistaken for lower volatility alternatives to broad semiconductor funds.
A Concentrated Global Market
The global memory market is heavily concentrated in three companies: Micron Technology (MU), Samsung Electronics (005930), and SK hynix (000660). Micron is readily available on a U.S. exchange, while Samsung and SK hynix primarily trade in South Korea. That can make direct ownership less straightforward for U.S. investors and has helped create a natural use case for ETFs.
Broad country funds offer one alternative. The iShares MSCI South Korea ETF (EWY), for example, provides exposure to Samsung and SK hynix, but it also holds financial, automobile, internet, and other South Korean companies. Dedicated memory ETFs offer a more targeted approach, although their portfolios can be highly concentrated in the same small group of manufacturers.
Memory ETFs Take Different Approaches
The Roundhill Memory ETF (DRAM), the first U.S.-listed ETF devoted specifically to memory stocks, launched on April 2, 2026. The actively managed fund invests in global companies tied to HBM, DRAM, NAND, solid-state drives, hard-disk drives, and other memory technologies. Its leading exposures include Micron, Samsung, and SK hynix, making it a relatively direct way to access the major global manufacturers.
DRAM’s early performance and asset growth demonstrated considerable investor interest, but they also highlighted the volatility of a concentrated theme. The fund fell approximately 32% in July after an unusually strong initial run. Investors nevertheless added $6.2 billion during the month, compared with $3.0 billion for the broader semiconductor fund VanEck Semiconductor ETF (SMH). That suggests many investors treated the selloff as an opportunity to increase exposure rather than exit the theme.
Other newer funds are attempting to differentiate themselves through portfolio construction:
The Kurv Memory Select ETF (KMEM) is even more concentrated in the three dominant producers. As of July 31, SK hynix, Micron, and Samsung represented approximately 85% of its look-through exposure. KMEM may appeal to investors seeking direct exposure to the leading manufacturers, but its results will also be heavily dependent on those three companies.
The Tema Memory ETF (DISK) takes a broader, actively managed approach, developed in partnership with semiconductor research firm SemiAnalysis. It can invest across HBM, DRAM, NAND, and other parts of the global memory market, including Asian companies that can be difficult for U.S. investors to access. Its active mandate also allows it to add newer entrants, such as Chinese memory producer ChangXin Memory Technologies (CXMT), following its public offering.
The Tuttle Capital Concentrated Memory Stack ETF (HBMX) extends beyond the largest memory manufacturers to include advanced packaging, testing, substrates, interconnects, and other technologies supporting memory production. The actively managed fund generally holds between 20 and 35 companies and requires meaningful memory-related revenue or strategic exposure. This may provide broader “picks-and-shovels” exposure, but it will not necessarily move in line with Samsung, SK hynix, or other major memory producers.
There are also more specialized structures. The Tuttle Capital Memory Stack Income Blast ETF (DRMP) combines memory exposure with an options strategy designed to generate income. Leveraged funds such as the Roundhill T-REX 2X Long DRAM Daily Target ETF (RAM) and the Defiance Daily Target 2X Long DRAM ETF (DRAL) are daily trading products rather than traditional long-term allocations.
The AI Bottleneck Is Not Limited to Memory
As detailed in an earlier research note, the AI trade is getting more granular, and it is not limited to memory. Photonics is also gaining attention, as investors look beyond the processors powering AI models to the technologies needed to connect them.
Photonics uses light instead of conventional electrical signals to move data. While processors provide the computing power, memory supplies the data, and photonics creates the network connecting the system. This becomes more important as data centers link larger numbers of AI chips and need to transmit greater volumes of information without consuming excessive power.
The photonics ETF lineup also offers several different approaches. The Tuttle Capital Pure Play Photonics ETF (FOTO) focuses on companies whose primary businesses are tied to photonics, including optical components, lasers, and data-center connectivity. It is the more concentrated choice for investors seeking direct exposure to the theme. Its holdings include companies such as Lumentum Holdings (LITE).
The Corgi Lithography & Semiconductor Photonics ETF (EUV) is a broader fund. In addition to optical networking and silicon photonics, the fund invests in lithography equipment, lasers, semiconductor inspection, sensing, and specialty materials. EUV may therefore behave more like a hybrid semiconductor-equipment and photonics strategy than a pure optical-connectivity fund.
The Tema Photonics & Optical ETF (LAZR) focuses primarily on companies enabling faster data movement among chips, servers, and data centers. LAZR takes a more global approach and holds several Asian and European optical-technology firms. Investors should note that LAZR also holds private-company exposure to Anthropic through a special-purpose vehicle, so its portfolio is not limited exclusively to publicly traded photonics companies.
Roundhill’s Photonics and Optics ETF (LYTE) is also in the filing pipeline, along with Aura’s AI Photonics ETF (PHOX). PHOX will so far be the only indexed product in the photonics space, which brings a different perspective among a field of active peers.
As with memory ETFs, the fund name alone does not tell the full story. Some photonics products emphasize smaller optical-component manufacturers, while others include large semiconductor companies, chipmaking equipment, private assets, or applications outside AI such as defense, medical imaging, and industrial manufacturing.
Bottom Line
Memory and photonics ETFs are generally more concentrated and potentially more volatile than broad semiconductor or technology funds. For many investors, they may make the most sense as satellite positions, alongside diversified equity and semiconductor exposure. These ETFs allow investors to express a specific view on where the next AI infrastructure bottleneck may emerge.
For more news, information, and analysis visit the Thematic Investing Content Hub.
