Tired AI Fears Sent This 6.9%-Paying Fund Soaring (But It's Still Cheap)

AI “bubble” fears pressured semiconductors, with the VanEck Semiconductor ETF (SMH) down 11% from its late-June peak, but the article argues the selloff is overdone. It highlights BlackRock’s BTX closed-end fund yielding 6.9% monthly and trading at an ~9.3% discount to NAV (discount narrowing from ~10.7% a week earlier), while positioning it for continued AI enterprise demand supported by rapid growth in OpenAI Codex users (5M→8M in ~1.5 months; 8M→15M in <1 month) and Anthropic revenue growth (from $787M to $11.5B, +1,400% YoY). BTX is cited as up 48% in 2026, recovering faster than the broader S&P 500 ETF as the discount compresses.

Analysis

The market is still treating AI spend as a binary bubble call, but the better framing is dispersion: front-end demand is holding while investors are repricing the slope of capex, not the destination. That favors names with durable share in the buildout, especially NVDA, while leaving higher-beta beneficiaries like MU more exposed to any temporary digestion if hyperscaler orders pause for even one quarter. In other words, the next leg is likely to be driven more by procurement cadence and utilization than by outright demand destruction.

For the CEF angle, the key mechanism is discount compression rather than pure NAV appreciation. If income buyers continue to rotate into AI exposure, closed-end funds holding crowded AI assets can outperform the underlying basket over the next 1-3 months simply because the market is paying less of a discount tax; that effect can reverse quickly if risk appetite fades or if the AI trade gets blamed for broader factor de-grossing. The real second-order winner is anything that can sell "AI exposure plus yield" to retail and advisory flows, which is a technical tailwind independent of near-term fundamentals.

Contrarian view: the consensus is likely overestimating how broad the next wave of AI monetization will be. Enterprise adoption is real, but the revenue pool is still concentrated in a small set of infrastructure vendors and model providers; if business users keep experimenting without scaling budgets, semis can underperform even while usage metrics look strong. The falsifier is a slowdown in enterprise budget growth or any guidance commentary implying that AI demand is moving from acceleration to normalization; in that case, MU would likely re-rate down faster than NVDA because memory is the cleaner cyclical read-through.

Key Decisions for Investors

• Overweight NVDA versus SMH on a 1-3 month basis: use pullbacks to add, with a thesis that share quality should protect it if AI demand merely slows rather than breaks. Falsify if hyperscaler capex commentary rolls over or NVDA forward estimates stop rising.
• Stay selective on MU: prefer a tactical long only if memory pricing and capex commentary confirm order durability; otherwise use MU as the higher-beta hedge against an AI digestion phase. Risk/reward is worse than NVDA if the market starts discounting a pause in orders.
• Use BTX/AI-yield closed-end funds as a flows trade, not a core fundamental bet: buy on discount widening, trim on discount compression. The edge is in NAV-to-price convergence over weeks, not years; fade if broader credit or rate volatility picks up.
• If you want a cleaner expression of 'AI demand is intact but crowded positioning is vulnerable,' consider a pair long NVDA / short SMH for 1-2 months. The idea is to isolate quality while hedging sector-level multiple compression; invalidate if the sector breadth improves and SMH regains leadership.
• No aggressive options expression unless volatility resets: implied vol is likely to overprice headline risk. Wait for a 5-10% sector pullback before buying upside in NVDA or SMH; if that pullback fails to materialize and AI spend data remains firm, the better trade is spot equity, not calls.