The big surprise for the second half could be the AI trade powering higher. Why it wouldn’t take much.
After a year of AI-driven gains, it’s tempting to assume the easy money has been made. Valuations have expanded, the market is crowded with AI narratives, and investors are on high alert for any sign that spending or monetization might disappoint. That setup, ironically, is why the path of least resistance in the second half may still be up. It wouldn’t take much—modest beats, small macro tailwinds, or incremental evidence of adoption—to extend the AI rally and broaden it beyond the narrow group of early leaders.
Expectations have cooled more than headlines suggest
Consensus has quietly drifted toward slower second-half growth in AI infrastructure spending. The logic is sound: supply constraints in advanced packaging and high-bandwidth memory (HBM), rising power bottlenecks, and the sheer base effect after a torrid first half. But that caution leaves room for upside. When expectations undershoot, a small “beat-and-raise” from one or two hyperscalers, or a single supplier guiding a touch higher on HBM or networking, can ripple across the stack. Because the AI value chain is tightly coupled, small changes at the top of the funnel often amplify downstream. A 5–10% increase in cloud AI capex can translate into much larger order upside for component specialists with operating leverage.
Positioning isn’t as stretched as it looks
Yes, flows chased the biggest winners. But beneath the surface, many “AI-adjacent” names—optical networking, power electronics, thermal solutions, specialty chemicals, niche semicap—have lagged the marquee chip designers. Volatility spikes and rotation into cyclicals washed out some momentum positioning. Systematic buyers and buyback programs are poised to re-engage if indices or leaders break to new highs. In a market where secular growth is scarce, even a whiff of renewed AI acceleration can pull in incremental capital quickly.
Rates don’t have to fall much to matter
Long-duration growth equities are highly sensitive to discount rates. It wouldn’t take a full-blown easing cycle to lift multiples; a 20–30 basis point slip in long yields or a single dovish signal on the policy path can support another turn of multiple expansion for quality AI assets, particularly those with visible earnings revisions. Meanwhile, a steady “soft-landing” backdrop—moderating inflation without a growth scare—keeps the cost of capital stable and risk premia contained. For AI leaders compounding revenues at 20–40% with rising gross margins, that environment remains fertile.
Monetization is edging from promise to line item
The first phase of the AI trade was infrastructure-led: build models, buy accelerators, stand up clusters. The second phase is about usage and monetization. Here, tiny changes in attach rates move mountains:
– Enterprise copilots: If only a small percentage of large-company seats convert to paid assistants at $20–$30 per user per month, that is billions in high-margin software revenue before considering upsell into workflow automation, security, or analytics.
– AI in vertical software: Price lifts of 5–10% tied to AI features can drop nearly dollar-for-dollar to gross profit at scale vendors.
– Developer productivity: If AI-assisted coding trims cycle time by even low double digits, CIO budgets can justify continued spend on inference and fine-tuning, sustaining cloud AI utilization.
None of these requires breakthrough AGI moments. They require procurement comfort, clearer ROI cases, and reference deployments—incremental steps already underway in many enterprises.
The stack is still broadening—and bottlenecks are investable
The narrative is shifting from “GPUs or bust” to a more diversified stack:
– Memory and packaging: HBM remains tight. A small upward revision in bit demand or pricing cascades into outsized earnings sensitivity for memory suppliers and advanced packaging tool vendors.
– Networking: As clusters scale, optical transceivers, switches, and interconnect standards become limiting factors. Modest upgrades in orders signal another leg of spend.
– Power and cooling: Datacenter power constraints are real—but that turns utilities, grid equipment, backup generation, and liquid cooling into beneficiaries. Power purchase agreements, substation buildouts, and upgraded thermal systems are multiyear programs, not one-offs.
– Edge and devices: The “AI PC” and on-device inference story is early. Even low single-digit unit share for AI-accelerated laptops and smartphones can catalyze a new replacement cycle and lift content per device.
– Services and integration: As pilots become production, systems integrators and consultants see higher utilization and better pricing, a late-cycle kicker.
Any single bottleneck easing, or proof that customers are paying through bottlenecks, extends the cycle without requiring heroic new products.
Operating leverage can do more of the lifting
Cloud platforms and leading software vendors now have two levers: top-line growth from AI services and margin expansion as utilization rises. Small improvements in GPU occupancy, inference efficiency, or model routing can add hundreds of basis points to gross margin on AI workloads. For suppliers, better yields in HBM or advanced packaging, and mix shifts to premium parts, drive operating leverage. The market tends to underprice these second-order effects until they appear in a quarter or two of results.
Technical and flow dynamics favor upside skew
– Buybacks: Many AI leaders are cash machines. If blackout periods end into stable macro data, repurchases can absorb supply and firm bids at higher levels.
– Passive flows: Index concentration means incremental dollars continue to overweight winners unless and until the earnings leadership changes.
– Options and volatility: Periods of elevated implied volatility have left dealers short gamma in some names; clean breakouts can force mechanical buying.
What could go wrong (and what would fix it)
– Power scarcity: Delays in grid interconnects or equipment lead times could slow deployments. The partial remedy is already in motion: multi-year PPAs, on-site generation, and diversified region buildouts.
– Slower enterprise adoption: If pilots stall, utilization lags. Watch for clearer ROI case studies and bundled pricing to nudge conversions.
– Regulation and antitrust: Headline risk can compress multiples temporarily, though near-term spending plans are unlikely to reverse.
– Supply chain hiccups: Export controls or packaging capacity could pinch. Diversification of suppliers and node transitions help, but timing matters.
– Macro shock: A growth scare would hit cyclicals and high-duration assets alike; conversely, mild disinflation or stable growth quickly reasserts the secular story.
Why it wouldn’t take much
– Low bar in places: Consensus assumes a deceleration. Slight upside to cloud capex or HBM shipments resets models.
– High earnings convexity: Small revenue beats in semis, memory, or networking can create outsized EPS revisions due to fixed-cost absorption.
– Scarcity premium: Few sectors offer multi-year double-digit growth with improving margins; incremental proof of durability is rewarded.
– Compounding evidence: Each incremental enterprise win, each percentage point of AI seat attach, each improvement in model efficiency reduces uncertainty and pushes the narrative forward.
Catalysts to watch in the second half
– Hyperscaler capex updates and comments on AI cluster utilization
– HBM pricing and capacity expansions; advanced packaging tool order books
– Networking lead times and 800G/1.6T transition milestones
– Utility and datacenter REIT disclosures on power projects and interconnect queues
– AI PC launch cycles and early attach data for on-device models
– Software pricing/packaging changes for AI features and early adoption metrics
– Macro prints that nudge rate expectations lower or confirm a soft landing
The second half doesn’t need fireworks to keep the AI trade powering higher. It needs a few incremental proofs that the flywheel—spend, build, use, monetize—is still turning and broadening. Given how tightly linked the stack is and how responsive flows can be, those proofs may not have to be large to move prices meaningfully.
This article is for informational purposes only and is not investment advice.
