Osterweis Capital Management Q2 2026 Equity Outlook

The five largest hyperscalers are projected to spend over $700 billion in aggregate capex this year, up more than 60% from 2025. The article frames this as a continuation of software and information services as quality compounders, with AI-driven infrastructure investment central to the outlook. The message is constructive for large-cap technology spend, though it is more thematic than event-driven.

Analysis

This level of hyperscaler spending is less about incremental model training and more about a multi-year infrastructure buildout that transfers pricing power from software vendors to the picks-and-shovels layer. The most durable beneficiaries are not just chip designers, but the bottleneck owners: advanced packaging, HBM memory, networking optics, power delivery, thermal management, and grid/interconnect equipment. In practice, the first-order demand hit is immediate, but the second-order earnings leverage likely shows up with a 6-18 month lag as supply chains tighten and utilization rates stay elevated.

The market is still underestimating how much this capex wave can crowd out capital efficiency narratives in software. If cloud providers keep reinvesting at this pace, near-term free cash flow expansion will be muted, which matters because those names have been priced as margin-stable compounders. That creates a subtle relative-value setup: long the enablers with visible order backlogs, short the parts of enterprise software most exposed to customer budget scrutiny if AI spend does not translate into quick monetization.

The main risk is not that AI demand disappears; it is that returns on capital stay uncertain long enough to force a digestion phase. If hyperscaler capex growth slows from triple digits to low double digits over the next 2-3 quarters, the entire trade can de-rate quickly because positioning is crowded and expectations are now anchored to sustained acceleration. A secondary risk is supply normalization in 2026: once lead times compress, margin expansion can shift from hardware vendors back toward the hyperscalers, so the trade is better expressed in nearer-dated catalysts than as a perpetual secular long.

Consensus is too focused on the narrative that all AI exposure is equivalent. The better trade is to separate cash conversion from revenue growth: companies selling constrained physical inputs are likely to enjoy more durable pricing than those selling aspirational software features. That gap is where the mispricing lives, especially if investors continue to value everything in the AI stack as if the same growth multiple is justified.

Key Decisions for Investors

• Overweight the AI infrastructure bottleneck basket over the platform names for the next 6-12 months: long semiconductor equipment, advanced packaging, memory, and networking exposure; these should capture the highest incremental margin from the capex surge with the least customer concentration risk.
• Pair trade: long hardware enablers / short high-multiple software beneficiaries that need rapid AI monetization to justify valuations. Enter on any 3-5% pullback in the software leg; target 10-15% relative underperformance over 2 quarters if capex stays elevated.
• Add a tactical long in power and grid-capacity beneficiaries for a 12-24 month horizon. The second-order bottleneck is electricity and cooling, and names tied to transformers, switchgear, and data-center power management should see sustained order growth.
• Use downside protection on hyperscaler-heavy indices via 3-6 month put spreads. If capex guidance inflects lower before revenue monetization catches up, the market will likely punish the whole AI complex first and ask questions later.
• Avoid chasing pure-play AI software at current levels unless there is clear usage-based monetization. The risk/reward is asymmetric to the downside if investors decide the capex boom is funding external suppliers rather than creating near-term platform profits.