Can AI Capex Extend The Semiconductor Cycle?

The key second-order effect is not just higher semiconductor demand, but a re-pricing of supply optionality. When hyperscalers keep spending, the scarce assets are no longer only GPUs; the bottlenecks migrate into advanced packaging, HBM, power delivery, and test equipment, which should widen the performance gap between “enabler” suppliers and the broader chip complex. That favors companies with exposure to the picks-and-shovels layer of the stack and penalizes any supplier dependent on cyclical handset/PC demand as the AI buildout monopolizes capacity.

The market is likely underestimating how sticky the margin expansion can be for analog and power-management vendors. In constrained environments, customers will pay for qualification certainty, design wins, and delivery reliability, which usually leads to longer-lived pricing power than in compute silicon itself. The trade implication is that earnings revisions may continue to drift upward for 2-3 quarters even if headline semiconductor revenue growth normalizes, because the mix shift toward higher-value content per system is still unfolding.

The main risk is that capex enthusiasm becomes self-correcting if hyperscalers hit ROI scrutiny in 2026, but that is a months-to-years issue, not a near-term one. Near term, the better reversal trigger would be a supply-side thaw: faster-than-expected capacity additions in foundry, HBM, or packaging could compress shortages and cap the multiple expansion in the most crowded names. Another contrarian angle is that the market may be over-owning obvious AI beneficiaries, while under-owning the less glamorous suppliers whose revenue is less headline-sensitive and whose free cash flow conversion is improving faster.

For the next 6-12 months, the setup argues for staying with the supply-constrained layer rather than chasing the most expensive AI compute leaders. If the cycle extends into 2027, the winners should be the firms that can monetize scarcity with the least incremental capital intensity, not those whose top line is largest. That makes relative value more attractive than outright beta exposure.