Why investors should stay long on AI stocks, according to Trivariate's Adam Parker

Trivariate Research founder Adam Parker argues the AI investment cycle may have 7–10 years to run and says productivity gains are still ahead, with fewer than 50% of the top 500 U.S. equities discussing AI-related productivity. Since May 2023, the S&P 500 has risen 84%, its technology sector 147%, SMH 335% and Nvidia 647%; in October 2026, SMH has fallen 11%, the S&P 500 technology sector 4.8% and Nvidia more than 5% amid borrowing concerns.

Analysis

The key distinction is between the duration of AI infrastructure spending and the duration of attractive equity returns: a multi-year buildout does not protect today’s valuations if funding costs rise, utilization disappoints, or customers demand faster payback. Credit stress is a transmission mechanism, not just a sentiment issue—if it raises the cost of financing data centers or weakens marginal buyers, orders can slow before end-user productivity appears in reported results.

The productivity thesis is plausible but not yet an earnings signal. Benefits may diffuse beyond infrastructure suppliers to adopters, while competition can pass some efficiency gains to customers rather than shareholders. Power availability, implementation costs, and the time needed to redesign workflows are potential bottlenecks. Thus, broad earnings revision breadth matters more than executive mentions of AI.

Near term (days to weeks), positioning and credit-sensitive selling can keep semiconductor volatility elevated. Over 1–3 months, watch earnings revisions and guidance across both infrastructure suppliers and non-tech adopters, alongside borrowing conditions for data-center projects. Over 6–18 months, realized labor-cost savings could support broader earnings, but a capex pause would expose concentrated supplier expectations. The bullish case is falsified by sustained capex-guide reductions, weakening AI-related orders, or no improvement in adopter margins/productivity despite continued spending.

Key Decisions for Investors

• Do not add to NVDA solely on the premise that the investment cycle has years to run; cycle length is not a valuation floor, and the supplied company-specific sentiment is neutral. Reassess against order trends, customer spending plans, and earnings revisions.
• Consider a small, staged relative-value position: long an equal-weight S&P 500 vehicle such as RSP versus short SMH, entering only after semiconductor relative weakness stabilizes or earnings breadth confirms diffusion beyond infrastructure. This expresses a broadening-productivity thesis while reducing dependence on the most crowded capex beneficiaries; it can lose if semiconductor earnings estimates keep rising faster than the broader market.
• Use the next 1–3 months of results as the confirmation window: track adopter margin guidance and labor-cost commentary, not AI mentions alone. If broad earnings revisions fail to improve while infrastructure spending remains high, reduce the broadening thesis rather than treating adoption narratives as proof of returns.
• Monitor credit conditions and project financing as early-warning indicators. A deterioration accompanied by data-center capex cuts or weaker supplier guidance would invalidate the long-cycle comfort argument and favor reducing semiconductor exposure; absent such confirmation, avoid treating a single month of weakness as evidence that the structural cycle is over.