Investors See AI Slowdown as a Temporary Step

BMO Capital Markets' Jennifer Lee said calls to slow AI development could create near-term market uncertainty, but AI infrastructure spending, investment and demand are unlikely to reverse. Concerns over employment displacement, data-center buildouts and profitability may moderate growth, while labor shortages and automation needs continue to support AI's long-term economic expansion.

Analysis

The relevant market distinction is between AI demand durability and the timing of capex conversion into earnings. Hyperscalers can sustain infrastructure spend despite public-policy friction, but investors are increasingly likely to penalize projects without measurable inference revenue or workload utilization. That favors the semiconductor and networking layers with near-term contracted demand—NVDA, AVGO, ANET and VRT—over software names whose valuations embed eventual AI monetization but face a longer and less visible sales cycle.

Labor scarcity makes automation spend less cyclical than the current debate implies, particularly in logistics, industrial automation and customer-service workflows. The second-order beneficiaries could therefore be ETN and PWR through power-grid and electrical buildout, plus ROK and TER where AI-driven automation raises ROI for customers; these exposures are less dependent on a single model provider's monetization. Conversely, data-center operators and power-intensive infrastructure names face a 6-18 month constraint from interconnection queues, power pricing and permitting, which can defer revenue even if end-demand remains intact.

Near term, policy rhetoric is more likely to widen dispersion than impair aggregate spending: a 1-3 month pause in enterprise pilots or restrictions on specific deployments would hit high-multiple application software first, while chip orders already inside the supply chain should prove more resilient. The contrarian risk is that infrastructure consensus has become too crowded: any downward revision to 2027 hyperscaler capex, evidence of lower GPU utilization, or a material improvement in AI-chip supply would compress NVDA/AVGO multiples before revenue declines. There is no standalone directional catalyst in this item; use it to refine exposure rather than add beta aggressively.

Key Decisions for Investors

• Maintain a 3-6 month quality AI-infrastructure basket long NVDA, AVGO, ANET and VRT, but fund it with an underweight in high-multiple application software such as C3.ai (AI) and UiPath (PATH). The intended payoff is relative: infrastructure estimates should remain better supported by backlog, while application multiples are more exposed to delayed monetization.
• Add a 6-18 month satellite long in ETN and PWR on pullbacks rather than chase semiconductor momentum. Grid equipment and engineering capture the power bottleneck created by data-center expansion; falsify if utility interconnection activity or large-load project announcements weaken materially over two consecutive quarters.
• Do not initiate a broad AI ETF long solely on this development. Set an alert around the next META, MSFT, GOOGL and AMZN capex updates: a collective reduction in forward capex guidance, explicit utilization concerns, or a 10%+ cut to NVIDIA next-twelve-month revenue estimates would warrant reducing the infrastructure basket.
• For downside protection around earnings, consider 3-6 month QQQ put spreads rather than single-name puts if AI exposure is already concentrated. This hedges a valuation-led technology de-rating while preserving idiosyncratic upside from infrastructure execution; reassess if real yields decline materially or hyperscaler capex guidance accelerates.