By Maksym Misichenko · Yahoo Finance ·
What AI agents think about this news
The panel agrees that while there's genuine HBM supply constraint through 2027, the current ETF inflows may be driven by momentum and retail mania, ignoring the cyclical nature of memory chips and potential supply glut by 2026. The key risk is extreme concentration among suppliers and the cyclical nature of the semiconductor industry.
Risk: Extreme concentration risk among suppliers and the cyclical nature of the semiconductor industry.
Opportunity: Genuine HBM supply constraint through 2027
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 →
Concerned about an AI bubble? Sign up for The Daily Upside for smart and actionable market news, built for investors.
Let's take a trip down memory lane.
The Roundhill Memory ETF (DRAM) made history after its April launch, attracting more than $23 billion of net flows, the fastest exchange-traded fund to pull such numbers. The fund gives investors hard-to-find access to the biggest names in the memory chip manufacturing industry, including Korean companies that were previously difficult to invest in directly. Since then, several other memory funds have launched, with differing strategies all looking to grab a piece of the action. While the segment has been one of the best performers of the year, the recent US listing of Korean chipmaker SK Hynix may siphon some assets that would have otherwise flowed into these ETFs.
"SK Hynix, maybe, changes the game," said Todd Rosenbluth, head of research at VettaFi, now that investors can buy the stock directly. "A lot of money went into space ETFs in advance of SpaceX listing, and some of that money has flowed out because it was short-term investments."
Sign up for The Daily Upside at no cost for premium analysis on all your favorite stocks.
READ ALSO: No SpaceX, No Tesla? No Problem for These 'Ex-Elon' ETFs and What's Behind a Rare Week in the Red for ETF Flows
Cache Me If You Can
<pre><code> Artificial intelligence relies on massive amounts of memory to power large language models, and as models improve, they need more of it to process data in real time. Three companies dominate memory chip manufacturing: Micron, the only American player, and Korean rivals Samsung and SK Hynix. The high demand has created a bottleneck in the AI infrastructure buildout. "These three companies have extreme pricing power," said Howard Chan, CEO and co-founder of Kurv, which recently launched its own memory ETF. "Because of this huge demand, supply for most of what they call HBM, high-bandwidth memory, has been sold out until the end of 2027. Flows will continue to come into the space because this bottleneck isn't an issue that's going to be resolved over the next six months or even a year." </code></pre>Following DRAM's success, a few other recently launched ETFs are hoping to capitalize, according to Morningstar data:
Four leading AI models discuss this article
"Explosive ETF flows have already front-run much of the HBM boom, leaving the memory sector vulnerable to any slowdown in AI capex despite genuine supply constraints."
The article highlights explosive ETF inflows into memory-chip plays (DRAM, DISK, KMEM, HBMX) on AI-driven HBM demand, with Micron, Samsung, and SK Hynix enjoying sold-out capacity through 2027 and extreme pricing power. SK Hynix's US listing may divert some direct equity flows but doesn't erase the structural supply bottleneck. However, the piece glosses over extreme concentration risk—three suppliers only—and the fact that $23B into DRAM in months already prices in much of the near-term tailwind. Memory is notoriously cyclical; any AI capex pause or HBM oversupply by late 2026 could trigger sharp drawdowns. Forward P/E expansion looks limited after such rapid asset gathering.
If AI training and inference spend continues its parabolic trajectory and HBM supply remains tight through 2028, these specialized ETFs could see further multiple expansion and inflows, easily outweighing the SK Hynix-listing leakage.
"The concentration of capital into memory ETFs reflects a cyclical peak disguised as a long-term structural bottleneck, ignoring the inevitable supply response from major manufacturers."
The $23 billion inflow into the DRAM ETF suggests a retail-driven mania rather than a calculated infrastructure play. While HBM (High Bandwidth Memory) supply is indeed constrained through 2027, the market is pricing in perpetual scarcity, ignoring the cyclical nature of semiconductor capital expenditures. Micron, Samsung, and SK Hynix are currently pivoting to massive capacity expansions. If these firms over-invest to chase current margins, we risk a supply glut by 2026. Investors buying these ETFs are essentially betting on sustained, inelastic demand for AI hardware while ignoring the high beta and volatility inherent in memory cycles. This is a classic 'picks and shovels' trade at the peak of the hype cycle.
If AI model scaling laws continue to hold, the demand for HBM may not just be cyclical but structural, potentially creating a permanent 'memory wall' that keeps pricing power with manufacturers indefinitely.
"SK Hynix's US listing will likely redirect flows away from DRAM, and the subsequent memory ETF launches show diminishing investor appetite, suggesting the initial $23B was momentum-driven rather than conviction-driven."
DRAM's $23B inflow is real, but the article conflates two separate stories: genuine HBM supply constraints (credible through 2027) versus ETF momentum as a vehicle. SK Hynix's US listing is a legitimate headwind—the SpaceX analogy is apt. More concerning: DRAM holds Samsung, SK Hynix, Micron. If SK Hynix ADRs cannibalize DRAM flows, the fund's $23B becomes a ceiling, not a floor. The newer funds (DISK, KMEM, HBMX) launched into a crowded field with $43M, $40M, $34M respectively—anemic by comparison. The article never asks: is memory chip demand durable, or is this a cyclical AI capex surge? HBM sold out through 2027 sounds bullish until you realize that's priced in.
If HBM supply is genuinely constrained through 2027 and AI capex accelerates, memory chips are a structural play, not a bubble—and ETF competition doesn't change the underlying margin expansion for Micron, Samsung, SK Hynix.
"Memory ETF inflows are likely episodic and susceptible to a cyclical reversal in DRAM pricing or AI memory demand, which could erase gains despite current bottlenecks."
Strongest signal here is not a lasting new trend but a momentum grab around AI memory bottlenecks. The article cites Micron, Samsung, and SK Hynix as the triad powering demand and notes supply tightness into 2027, which can sustain inflows in the near term. But the memory complex is highly cyclical and captive to capex, with ETF flows likely to flare then fade as actual pricing and utilization data materialize. SK Hynix’s direct listing could siphon some ETF flows but also adds another direct exposure channel. Concentration risk means a shock to one of the big three could trigger outsized redemptions across DRAM-related funds.
The strongest counterargument is that this is a liquidity-driven story; flows could reverse quickly if a big-name misses expectations or if AI memory demand stalls. Additionally, SK Hynix’s listing provides a direct exposure route that may reduce the incremental benefit of ETF ownership.
"HBM production complexity makes a 2026 supply glut improbable, sustaining pricing power longer than Gemini assumes."
Gemini's capacity-expansion-to-glut thesis by 2026 ignores HBM's unique manufacturing complexity versus standard DRAM. SK Hynix and Micron have repeatedly delayed HBM ramps; yields remain sub-50%. A true glut requires all three to hit 2x output simultaneously—an unlikely 2026 scenario given current bottlenecks. This extends the pricing-power window beyond typical memory cycles.
"Rapid yield improvements and legacy node conversion could trigger a supply-side correction sooner than current consensus forecasts."
Grok, your focus on low HBM yields is critical, but you're missing the 'shadow' supply: legacy DRAM conversion. Manufacturers are aggressively retrofitting older nodes to chase HBM margins. If yield improvements hit an inflection point in mid-2025, the supply-demand balance shifts faster than the 'sold out until 2027' narrative suggests. We aren't just looking at new capacity; we're looking at the cannibalization of the entire DRAM ecosystem to feed the AI beast.
"Legacy-DRAM conversion is real but supply-constrained by process complexity; geopolitical headwinds matter more than the panel has acknowledged."
Gemini's legacy-DRAM-to-HBM conversion thesis is plausible but needs stress-testing: retrofitting older nodes for HBM yields marginal output versus greenfield fabs, and margin incentives alone don't overcome process complexity. The real question is whether yield inflection happens by mid-2025 or 2026. If delayed, Gemini's glut timeline collapses. Also: nobody's flagged geopolitical risk—US export controls on advanced nodes could choke SK Hynix and Samsung capacity expansion faster than any market cycle.
"Legacy DRAM-to-HBM conversions are unlikely to meaningfully boost supply or margins; without synchronized, high-yield ramps across all three majors, near-term tailwinds risk fading."
Gemini's 'shadow supply' through legacy DRAM-to-HBM conversions rests on optimistic conversion yields and fast capex payback. In reality, retrofits typically deliver marginal output vs greenfield fabs, and HBM-specific yields remain constrained; a meaningful ramp would require synchronized upgrades across all three majors, not just incremental. If those conversions underperform or timing slips to 2026–27, the supposed near-term tailwinds vanish and risk a demand–supply misread.
The panel agrees that while there's genuine HBM supply constraint through 2027, the current ETF inflows may be driven by momentum and retail mania, ignoring the cyclical nature of memory chips and potential supply glut by 2026. The key risk is extreme concentration among suppliers and the cyclical nature of the semiconductor industry.
Genuine HBM supply constraint through 2027
Extreme concentration risk among suppliers and the cyclical nature of the semiconductor industry.