This AI ETF Shows Why Your AI Portfolio Isn't Making You Money

The Themes Generative Artificial Intelligence ETF (WISE) has lagged the leading AI beneficiaries, down about 4% YTD versus Nvidia up 18% and AMD up 79% YTD, while carrying a 35 bps fee and roughly $31 million in AUM. The article argues that equal-weight exposure to unprofitable AI startups dilutes returns versus direct ownership of hyperscaler and hardware capex winners such as NVIDIA, AMD, and QQQ. It frames the 2026 AI trade as a hardware and infrastructure story rather than a broad thematic basket.

Analysis

The market is converging on a cleaner winner-take-most structure in AI: compute, networking, memory, and power delivery are compounding off hyperscaler capex, while the long tail of “AI software” is starting to behave like a financing story rather than a product story. That matters because passive or quasi-equal-weight exposure systematically underowns the names with accelerating free cash flow and overowns the names that need repeated capital raises to survive. In other words, the ETF is not just lagging performance; it is structurally short the operating leverage that is becoming the dominant source of AI returns.

The second-order effect is that capital allocation is likely to become even more concentrated. If the market keeps rewarding the infrastructure layer, suppliers with real bottlenecks — advanced packaging, HBM memory, optical interconnects, server OEMs, and power/thermal components — should see better pricing power than the article’s focal names imply. That leaves smaller AI software names vulnerable to a double hit: slower multiple expansion and higher cost of capital as investors demand proof of distribution, not just model access.

The main risk to this view is timing. In the next few weeks, the crowd can still bid the “AI beta” sleeve on any incremental guidance beat or model launch, but over a 3-6 month horizon, cash flow should reassert itself. The countertrend catalyst would be a broad enterprise software re-acceleration or a regulatory shock to hyperscaler capex, either of which would temporarily revive the thematic basket’s diversification premium. Absent that, the discount rate on unprofitable AI names should keep rising.

Consensus is probably still underestimating how much of AI value creation is being captured outside the obvious ticker set. The market talks about AI as a software revolution, but the monetization engine is looking more like a supply-chain industrial cycle with software optionality attached. That makes direct ownership of the bottleneck beneficiaries superior to owning the “idea” of AI through a diluted basket.

Key Decisions for Investors

• Long NVDA / short WISE: express the view that capex concentration will keep outperforming the diversified thematic basket over the next 3-6 months; target 15-25% relative outperformance if hyperscaler spending remains intact.
• Long AMD / short BBAI: pair the highest-quality beneficiary against a cash-burning small-cap laggard; use as a cleaner way to isolate valuation dispersion in AI hardware vs AI narrative names.
• Add MSFT on 2-4 week weakness, but size it as a defensive AI compounder rather than a hypergrowth trade; upside depends on Azure monetization re-accelerating, downside is capped by balance sheet and recurring cash flow.
• Avoid or underweight QUIK and other tiny AI adjacency names until they show at least two consecutive quarters of revenue acceleration and positive gross profit leverage; these are the most likely to suffer survivorship risk if the theme cools.
• For aggressive accounts, buy 3-6 month call spreads on NVDA or AMD and fund them by selling thematic AI ETF call spreads; this captures the dispersion trade while limiting exposure to a broad de-rating in AI sentiment.