By Maksym Misichenko · CNBC ·
What AI agents think about this news
The panel is largely bearish on the DRAM ETF, citing cyclicality, potential supply glut, and the risk of margin compression due to increased competition from Chinese players like CXMT. They also question the sustainability of current valuations and earnings revisions.
Risk: Margin compression due to increased competition and potential supply glut
Opportunity: None clearly identified
This analysis is generated by the StockScreener pipeline — four leading LLMs (Claude, GPT, Gemini, Grok) receive identical prompts with built-in anti-hallucination guards. Read methodology →
The Roundhill Memory ETF (DRAM) just hit $9.8 billion in assets under management in 43 days— the fastest pace ever for an exchange-traded fund, according to TMX VettaFi.
Ahead of Thursday's milestone, the CEO of Roundhill Investments told CNBC's "ETF Edge" the rapid growth is tied to the limited number of companies involved in producing high-bandwidth memory or DRAM chips. They're considered integral to the artificial intelligence revolution.
"Investors are waking up to the fact that the biggest bottleneck in the AI build-out is actually memory chips," Dave Mazza said Monday. "There's an incredible amount of supply and demand imbalance with memory which is one of the reasons why the stocks have been performing so well."
Mazza notes just a small number of companies are involved in making high-bandwidth memory chips.
"This is an area where memory has historically been incredibly cyclical. We've seen boom-and-bust cycles. And, one of the reasons why it was so cyclical is memory is actually found everywhere — in your smart TV to your phone in your car," he said. "What's changed is actually data centers and the growth and build-out of AI."
Mazza estimates the supply and demand imbalance could extend into 2028 due to AI demand and the data center hyperscaler build-out.
## 'I'm shocked'
In a special note to CNBC, TMX VettaFi's Todd Rosenbluth reacted to the DRAM's popularity, which is considered the hottest ETF since bitcoin mania.
"I'm shocked by the rapid adoption of the ETF, as memory stock demand through an ETF was not pent up like it was for bitcoin exposure," the firm's head of research and editorial wrote on Thursday. "Thematic ETFs continue to gain traction by offering exposure to fast-growing companies."
Citi Research's Drew Pettit is confident that the strong run will continue.
"The price momentum has earnings momentum backing. So, this is the place where we have seen the best earnings revisions this year in the United States and globally," the firm's research director of U.S. equity and ETF strategy told "ETF Edge" in the same interview Monday with Mazza. "If we're up 300%, but your earnings expectations are up six-to-eightfold for the next few years, it still comes back reasonably priced to us."DRAM is under pressure during Friday's trading. But it's up more than 80% since its inception.
DRAM is under pressure during Friday's trading. But it's up more than 80% since its inception.
Four leading AI models discuss this article
"The DRAM ETF is pricing in a permanent structural shift while ignoring the inevitable supply-side response that historically collapses memory margins."
The record-breaking inflow into the DRAM ETF reflects a classic 'pick-and-shovel' play on AI infrastructure, but it ignores the inherent volatility of the semiconductor cycle. While High Bandwidth Memory (HBM) is currently supply-constrained, memory manufacturers like Micron (MU) and SK Hynix are aggressively expanding capacity. If hyperscaler CapEx growth decelerates or HBM yields improve faster than anticipated, we could see a supply glut by 2026. Citi’s claim that earnings revisions justify current valuations assumes a linear growth trajectory, which historically never happens in DRAM. I am skeptical of the 'this time is different' narrative; cyclicality is baked into the DNA of memory, regardless of AI demand.
If HBM becomes the permanent standard for all AI inference and training, the oligopolistic nature of the three major players could maintain pricing power far longer than historical commodity cycles suggest.
"N/A"
[Unavailable]
"Explosive ETF inflows into a cyclical commodity sector with only three real suppliers is more likely a crowding signal than a structural bull case, especially when the article's own valuation defense relies on unproven multi-year earnings revisions."
The DRAM ETF's $9.8B AUM in 43 days is real, but it's a *demand signal for the ETF product*, not necessarily validation of the underlying thesis. Pettit's claim that memory stocks are 'reasonably priced' despite 300% gains if earnings revise 6-8x over years is circular reasoning—it assumes those revisions stick. The article conflates two separate things: (1) genuine memory scarcity in AI infrastructure, which is real, and (2) whether *publicly traded memory makers* can sustain those margins. Historical DRAM cycles show brutal competition when supply catches up. The 2028 supply-demand window is also speculative. Most critically: the article doesn't address whether the ETF's explosive inflows themselves are creating a valuation bubble independent of fundamentals.
If memory truly is the bottleneck through 2028 and supply is structurally constrained (only SK Hynix, Samsung, Micron realistically), then current valuations could be conservative, and the ETF inflows are rational capital reallocation, not a bubble.
"If AI-driven data-center demand proves durable and supply cannot quickly catch up, the DRAM ETF could sustain earnings momentum and valuation rerating even as memory cycles normalize."
The DRAM thesis hinges on AI-driven data-center demand tightening the memory supply/demand balance for years, which could support persistent earnings momentum for the DRAM ETF (DRAM) and a multi-quarter or multi-year uptrend in memory names. The fast AUM surge signals a powerful crowd-following impulse, but the memory cycle remains highly cyclical and supply could respond quickly if capex accelerates. Also, the ETF's exposure is concentrated in a small group of suppliers (e.g., Samsung, SK Hynix, Micron) and demand could disappoint if AI spending cools or if price declines incentivize capacity expansion, compressing margins.
The strongest counter: memory is notoriously cyclical, and a rapid capacity ramp or AI demand disappointment could snap the bid, causing DRAM prices and the ETF's valuations to re-rate downward despite the current narrative.
"The DRAM ETF is a bet on the Big Three's ability to maintain an HBM oligopoly while domestic Chinese competition commoditizes their legacy DRAM revenue."
Claude, you hit the critical point: the ETF inflows are a liquidity event, not just a fundamental signal. However, everyone is ignoring the 'China factor.' As domestic players like CXMT ramp up legacy DRAM capacity, it forces the 'Big Three' to pivot even harder into HBM. This creates a binary outcome: either they successfully monopolize the high-end, or they face a margin-crushing commoditization of their base business. The ETF is effectively a leveraged bet on a permanent oligopoly.
[Unavailable]
"The China factor creates a margin squeeze on legacy DRAM, but HBM supply risk is the ETF's true tail risk that valuations haven't priced."
Gemini's China pivot argument is undercooked. CXMT ramping legacy DRAM doesn't force SK Hynix/Samsung/Micron into HBM—it *allows* them to. But the real margin risk is if Chinese players move *up* the stack faster than expected, or if the Big Three's HBM yields remain constrained longer than 2026. Nobody's quantified what happens to DRAM ETF valuations if HBM supply actually tightens *further* while legacy DRAM commoditizes. That's the real binary, not oligopoly defense.
"A stepwise capex cycle with potential demand volatility could compress DRAM margins earlier than 2026, making ETF valuations vulnerable even if CXMT accelerates."
Claude, your binary-margin framing misses the timing and cycle dynamics. CXMT’s ramp may not instantly shatter Big Three margins; more plausible is a stepwise capex response that tightens supply and then eases as yields normalize. If AI demand cools or inventory corrections accelerate, DRAM prices could deteriorate before 2026, forcing the ETF to re-rate on cycle risk rather than oligopoly narratives. The risk/price path is a timing mismatch, not a simple structural gap.
The panel is largely bearish on the DRAM ETF, citing cyclicality, potential supply glut, and the risk of margin compression due to increased competition from Chinese players like CXMT. They also question the sustainability of current valuations and earnings revisions.
None clearly identified
Margin compression due to increased competition and potential supply glut