The Roundhill Memory ETF drew $1.113 billion in net daily inflows, one of the largest single-day creations across the U.S. ETF market, as investors continued to crowd into a concentrated trade on artificial-intelligence memory suppliers.

ETF.com’s daily fund flow report, published May 8 and covering May 7 trading activity, listed the Roundhill Memory ETF, ticker DRAM, as the second-largest ETF creation of the day. Only the Vanguard S&P 500 ETF attracted more capital, with $1.460 billion in inflows. DRAM’s one-day intake lifted its assets under management to $5.034 billion and represented a 22.11% change in assets, according to ETF.com’s flow table.

The fund’s asset surge is striking because DRAM began trading only on April 2, giving it a little more than a month in the market before crossing the $5 billion threshold. ETF.com separately reported that the fund reached that level in 25 trading days, one trading day longer than the iShares Bitcoin Trust ETF took to reach the same milestone after spot bitcoin ETFs launched in January 2024. Unlike IBIT, which entered a newly approved category alongside multiple competing funds, DRAM has so far operated as the only dedicated memory-themed ETF on the market.

The flows show how quickly investors have moved from a broad artificial-intelligence equity theme into more specific hardware bottleneck exposures. DRAM is built around companies producing and supplying high-bandwidth memory, NAND and DRAM. Roundhill describes computer memory and storage as a secular growth area tied to the multiyear buildout of AI infrastructure, positioning the ETF as a pure-play vehicle rather than a broad semiconductor allocation.

The distinction mattered in the May 7 flow data. The VanEck Semiconductor ETF, one of the most established U.S.-listed semiconductor ETFs, recorded the largest redemption across all ETFs, losing $1.319 billion. At the same time, DRAM attracted more than $1.1 billion. That combination suggests the day’s activity was not simply a general risk-on move into chip exposure. Investors appeared to be narrowing allocations toward the memory segment while taking money out of broader semiconductor baskets.

Other major creations on the day were concentrated in large liquid equity and sector funds. First Trust Dorsey Wright Focus 5 ETF added $842.29 million, iShares Core S&P 500 ETF added $626.95 million, SPDR S&P 500 ETF Trust added $440.50 million, Industrial Select Sector SPDR Fund added $406.84 million, and Technology Select Sector SPDR Fund added $331.51 million. DRAM’s place near the top of that list put a young thematic ETF alongside the largest and most liquid core portfolio instruments in the market.

The fund’s rapid growth reflects the market’s reassessment of memory chips in the AI supply chain. High-bandwidth memory is used in advanced AI accelerators because it can move large amounts of data at high speeds while operating close to graphics processors and custom AI chips. As hyperscale cloud providers and AI developers expand computing capacity, investors have increasingly treated memory suppliers as strategic infrastructure companies rather than commodity component vendors.

ETF.com reported that SK Hynix, Samsung Electronics and Micron Technology together account for roughly three-quarters of DRAM’s portfolio. That concentration gives the ETF direct exposure to the dominant global suppliers of high-bandwidth memory, but it also makes the fund unusually dependent on a small number of companies and on the durability of the current memory upcycle. Smaller positions in related storage and memory names have also participated in the rally, but the core trade remains tied to the leading HBM producers.

Roundhill’s fund page identifies Samsung, SK Hynix, Micron, Kioxia and SanDisk among the fund’s top holdings as of the launch period. The ETF has a 0.65% expense ratio, trades on Cboe BZX and is actively managed, although Roundhill says turnover is generally expected to be limited to quarterly rebalancing. The fund also uses total return swaps in calculating some exposure weights, according to its holdings disclosure language.

The ETF’s rise has been amplified by performance as well as creations. ETF.com reported that DRAM was up about 88% since launch as of May 8 and was trading higher again during the Friday session. That performance, combined with the scale of the latest inflow, likely pushed assets even higher after the published May 7 snapshot. In ETF markets, strong performance can mechanically lift assets, while new share creation brings in additional capital. DRAM has benefited from both forces in compressed time.

For ETF issuers, the development is a reminder that the thematic-fund cycle remains alive even as investors have favored low-cost core index products and active ETFs in other parts of the market. A fund can scale rapidly when it offers exposure that is difficult to replicate through traditional U.S. equity portfolios. DRAM also provides U.S. investors with a packaged route into non-U.S. memory leaders such as Samsung and SK Hynix, which are central to the AI memory supply chain but are not as straightforward for many U.S. investors to trade directly as domestic large-cap technology stocks.

The flow pattern also points to a changing role for ETF products in thematic allocation. Earlier AI trades often centered on broad semiconductor ETFs, software funds, robotics funds or single-stock exposure to dominant accelerator makers. DRAM’s growth shows that investors are increasingly trying to isolate specific components of the AI capital expenditure cycle, including memory, advanced packaging, power infrastructure, data-center equipment and cooling. In that environment, fund flows can become a real-time barometer of which part of the AI stack investors believe is facing the tightest supply-demand balance.

Recent industry commentary supports that framing. Micron said in March that its stronger results and outlook reflected memory demand driven by AI, structural supply constraints and company execution. Reuters reported on May 7 that SK Hynix had received unusual offers from large technology customers seeking to secure memory-chip supply, including proposals linked to production expansion and equipment. Those reports align with the investor narrative that memory capacity is not expanding quickly enough to meet near-term demand from AI infrastructure buyers.

Still, the ETF’s speed of growth brings investment risks that are distinct from the AI demand story. DRAM is a narrowly focused fund with heavy exposure to a small number of companies in a historically cyclical industry. Memory markets have often moved through sharp boom-and-bust cycles as manufacturers add capacity during periods of strong pricing, only to face oversupply when demand normalizes or new production comes online. The current bull case argues that AI has changed the duration and quality of demand. The bear case is that the industry’s economics have not been repealed, only delayed.

The concentration issue is especially important for allocators using DRAM as part of a diversified portfolio. A broad semiconductor ETF may hold dozens of companies across foundries, chip-design firms, equipment makers, analog suppliers and memory producers. DRAM intentionally narrows that exposure. That precision may be valuable for investors seeking a pure memory allocation, but it also leaves less room for offsetting performance if the memory trade reverses, if HBM pricing expectations cool, or if the largest holdings face execution issues.

Liquidity and trading behavior will also be watched closely because the fund has scaled so rapidly. Large inflows can improve secondary-market trading depth and visibility, but they also require efficient creation and redemption activity in the underlying portfolio or related instruments. When a young thematic ETF experiences unusually large daily flows, market makers, authorized participants and the issuer must manage execution across a portfolio that includes international equities and, in some cases, derivative exposure. The ETF wrapper can make access easier for end investors, but it does not eliminate the underlying market structure considerations.

DRAM’s rise also has competitive implications. ETF.com noted that rivals are expected to come to market, and the asset growth in DRAM is likely to accelerate interest from other issuers seeking AI supply-chain niches. Once a theme demonstrates multibillion-dollar demand, competitors often respond with variations on cost, index methodology, leverage, options income or active stock selection. For now, DRAM’s first-mover advantage has been decisive, but the category may become more crowded if the AI memory trade remains in focus.

The daily flow data also captures a broader allocation tension. Investors continue to add large sums to core index ETFs such as VOO, IVV and SPY, reinforcing the dominance of cheap beta products. Yet at the same time, they are willing to place large tactical allocations into narrow thematic ETFs when a market narrative gains urgency. The result is a barbell pattern in ETF demand: massive core holdings at one end and high-conviction satellite themes at the other.

For advisers and institutional users, the question is whether DRAM should be treated as a tactical trade, a long-term AI infrastructure sleeve or a high-volatility sector bet. The answer depends on portfolio construction. Investors who already hold semiconductor funds, technology sector ETFs or individual chip stocks may have more memory exposure than they realize. Adding DRAM on top of those holdings could increase exposure to the same AI capital spending cycle and to the same valuation assumptions embedded in chip-related equities.

The May 7 flows nevertheless show that the ETF market is assigning significant value to specificity. DRAM’s $1.1 billion creation was not just large for a young fund; it was large relative to the entire ETF universe on that day. The fund’s ascent from launch to more than $5 billion in assets in roughly five weeks reflects a convergence of product design, market timing and investor demand for tradable AI bottleneck exposure.

The next test will be whether flows remain durable after the initial performance surge. The most successful thematic ETFs typically need more than early momentum: they require sustained liquidity, a clear use case in portfolios and an investable universe that can support large assets without excessive concentration problems. DRAM has already shown there is demand for a memory-specific ETF. The durability of that demand will depend on whether the AI memory cycle continues to deliver earnings growth strong enough to justify the speed and scale of the capital now entering the fund.