DRAM Powers Another Big Week of ETF Inflows

US-listed ETFs took in $39.6 billion last week, led by SPY at $4.2 billion, VOO at $2.7 billion, and an outsized $2.7 billion into Roundhill Memory ETF (DRAM), which now has nearly $6 billion in AUM and has roughly doubled since debut. On the outflows side, QQQ lost $3.8 billion, SMH shed $2.2 billion, and SOXL lost $964 million, suggesting some profit-taking in crowded tech and semiconductor trades. The week also saw $907.9 million into IBIT as bitcoin rebounded above $80,000, while ongoing US-Iran tensions remained a background risk.

Analysis

The flow tape is telling us this is no longer just an AI capex trade; it is becoming a crowded expression of scarcity within the AI supply chain. When a niche thematic vehicle absorbs low-single-digit billions in a week, it usually means marginal buyers are chasing a narrative with limited float and high reflexivity, which can keep the basket extended for weeks but also makes it fragile if momentum cools. The bigger implication is that capital is being pulled out of broader AI proxies into the most levered upstream beneficiaries, so the next leg may be driven less by fundamentals and more by positioning feedback loops.

The rotation out of large-cap growth and semis looks like profit-taking, but the second-order effect is more important: it reduces index-level breadth even as headline indices hold up. That tends to create a market where the benchmark can stay resilient while under the hood leadership narrows, increasing the odds of a sharp factor unwind if rates tick higher or a single AI capex guide disappoints. The presence of strong inflows into cash-like T-bills alongside risk assets suggests investors are simultaneously adding beta and keeping dry powder, which is consistent with a tactical risk-on stance rather than a durable all-in commitment.

Crypto inflows fit the same pattern: investors are buying a beta expression on the same liquidity/animal-spirits trade rather than making a strong fundamental call on adoption. If Bitcoin fails to hold recent breakout levels, these inflows can reverse quickly because the marginal buyer is likely momentum-sensitive. The key risk horizon is days to a few weeks, not quarters; the trade can persist if prices remain orderly, but it is highly dependent on continued index strength and no negative surprise in AI earnings or geopolitics.

The contrarian read is that the market may be overpaying for the most obvious beneficiaries of the AI buildout while underestimating downstream pressure on margins for users of semis and memory over time. If memory pricing normalizes, the current inflow surge can compress into a crowded-to-crowded unwind even if AI demand remains healthy, because the ETF wrapper accelerates narrative saturation faster than underlying supply chains can re-rate. In other words, the opportunity may be less in chasing the winner and more in fading the second-order beneficiaries once ownership gets too concentrated.

Key Decisions for Investors

• Fade crowded momentum tactically: short QQQ via 1-2 week put spreads or a small outright short against a long cash/T-bill sleeve if breadth continues to narrow; target a pullback of 2-4% with tight risk to recent highs.
• Pair trade: long DRAM / short SMH for 2-6 weeks, betting that memory scarcity stays the cleanest AI beneficiary while broad semis give back some of the recent profit-taking rotation; use a 1:1 or 1:1.5 notional ratio and stop if SMH reclaims relative strength.
• Buy downside protection on IBIT for the next 30 days with out-of-the-money puts or put spreads; crypto inflows are momentum-dependent and can reverse sharply if BTC loses the recent breakout zone.
• Add SGOV on weakness as a financing sleeve for risk trades; the inflow pattern suggests investors want dry powder, and holding that optionality is preferable to chasing the most extended beta at this stage.
• If willing to express a broader factor view, run a short QQQ / long XLF or cash-like carry overlay for 1-3 months to monetize narrowing leadership, but only if rates stabilize and earnings revisions stop broadening upward.