Three under-the-radar growth stocks that are powering the AI supercycle

The AI supercycle is shifting from chip-driven speculation to an execution phase constrained by physical infrastructure and heavy capex. Demand for foundational AI infrastructure continues to outpace supply, implying ongoing spending pressure across the ecosystem rather than a near-term demand slowdown. The piece is broadly constructive for long-term AI investment, but it mainly signals capacity and monetization bottlenecks rather than a discrete catalyst.

Analysis

The market is moving from valuing AI as an IP story to valuing it as a utilities-and-logistics story. That shift re-rates the bottleneck owners: power interconnects, transformers, liquid cooling, fiber, networking, and grid equipment should compound faster than the headline semis trade because their order books are tied to multi-year buildouts, not next-quarter model demand. The second-order winner is the “boring” capex stack with pricing power and long lead times; the loser is anything exposed to AI optimism without direct access to scarce physical inputs.

A key underappreciated effect is that scarcity migrates from GPUs to everything around them, which compresses margins for integrators and lowers elasticity for hyperscalers. If compute growth is now gated by power availability and data center fit-out, then the next phase of outperformance should cluster around firms with capacity reservations, backlog visibility, and balance-sheet strength to pre-buy inventory. Smaller vendors may see revenue acceleration, but they are also the most exposed to working-capital strain, execution slippage, and customer concentration if the build cycle pauses.

The main risk is not AI demand rolling over; it is a temporary capex digestion phase where hyperscalers slow bookings while they absorb prior spend. That would hit the more momentum-sensitive names first, while infrastructure suppliers with contracts already in hand lag the slowdown by 2-4 quarters. Conversely, any meaningful easing in power, export controls, or supply-chain lead times would quickly flatten the bottleneck premium and rotate leadership back toward the semiconductor complex.

Consensus is likely still too focused on model-level monetization and not enough on the physical conversion rate of dollars spent into usable compute. The move is probably underdone in infrastructure beneficiaries because investors still anchor on chip margins, but overdone in names that benefit only from AI narrative rather than direct capex capture. The trade is to own the constraint, not the concept.

Key Decisions for Investors

• Long a basket of AI infrastructure enablers for 6-12 months (e.g., VRT, ETN, ANET, DLR) vs. short a high-beta AI software basket; thesis: physical bottlenecks should sustain pricing power in the build-out layer even if model monetization stays uneven.
• Buy pullbacks in semicap equipment only on evidence of backlog extension; otherwise favor a pairs trade long infrastructure/utility-capex names vs. short a basket of legacy industrials with no AI exposure. Risk/reward: 2-3x upside if capex remains elevated, but tight stop if hyperscaler spend guides down for two consecutive quarters.
• Use options to express convexity in power-grid beneficiaries: call spreads in ETN or VRT over 6-9 months. The catalyst is order-book conversion from data-center electrification; downside is limited to premium if permitting or interconnect delays push out bookings.
• Short the most narrative-dependent AI beneficiaries with weak cash-flow visibility on any capex digestion signal from hyperscaler earnings. Time horizon 1-3 months; target a 15-20% drawdown if bookings slow while infrastructure names hold up.