Analysis

The market is entering a second-order AI capex trade: the first winners were compute and leading-edge semis, but incremental spend is now drifting toward the least obvious bottlenecks — packaging, substrate, power delivery, thermal management, and server assembly. That broadens the alpha opportunity set because these businesses often have lower valuation starting points, less direct exposure to GPU cycle volatility, and more durable backlog conversion as datacenter buildouts extend over multiple years rather than quarters. TSM remains the cleanest way to express continued AI intensity, but near-term upside is increasingly about mix and pricing power rather than pure unit growth. The more interesting setup is in the “picks-and-shovels of picks-and-shovels,” where earnings revisions can lag order momentum by 1-2 quarters; that delay creates a window for rerating before consensus catches up. NVDA is still the sentiment anchor, but if capex broadens, its relative outperformance may moderate as the market rotates into infrastructure beneficiaries with less crowded ownership. The main risk is not demand collapse but a capex digestion phase: after a period of aggressive pre-buying, investors could see a 1-2 quarter air pocket if hyperscaler procurement timing slips or financing conditions tighten for the next wave of private AI spend. A sharper risk is that the market has already begun to discount “AI everywhere,” so names with the most direct narrative may underperform if the trade becomes about operating leverage and balance-sheet quality instead of story beta. The contrarian angle is that energy and grid constraints may be the true binding constraint, which would favor power equipment and utility-linked exposures over pure semiconductor beta. We would treat any pullback in TSM as a buyable consolidation, but prefer expressing the broader theme through a basket of lower-valuation AI infrastructure suppliers where estimate revisions can drive multiple expansion. The next leg higher likely comes from earnings confirmation in adjacent supply chain names, not from another headline about model training budgets.