Morgan Stanley: AI Investment Enters a New Phase of Barbell Allocation
nashnova research
Morgan Stanley, drawing on an AI stock-mapping database covering roughly 3,600 global equities, argues AI investing has moved past the pure upstream-compute bet into a barbell strategy — holding infrastructure bottlenecks on one end while adding software and AI-adoption plays on the other.
What is the "barbell strategy," and why now?
The barbell — concentrating positions at two extremes while leaving the middle light — rests on one core call: compute infrastructure and early-stage software / AI-adoption companies will generate alpha simultaneously, not in the sequential sector rotation of past cycles.
This means → Morgan Stanley sees the AI trade as no longer linear; both ends move at once.
Since this AI cycle began, semis and infrastructure have outperformed the S&P 500 by 500–700 basis points. The bank says it is now time to increase exposure to software and adoption names.
Why can the compute boom last this long?
Three constraints are choking supply expansion: labour shortages, a power gap, and regulatory approvals. The report estimates a 57 GW data-centre power shortfall globally from 2026 to 2028.
Hyperscaler capex growth is projected to slow from 93% in 2026 to 14% in 2028; the market is gradually pricing in a hardware-cycle peak.
In plain terms = demand is not fading — there is simply not enough power, not enough workers, and not enough permits. Supply is pinned down, and that passively extends the boom in bottleneck segments.
The bank flags an upside risk too: if demand keeps beating expectations and bottlenecks persist, the semi/infra rally could outlast consensus by a wide margin.
Has the adoption side hit an earnings inflection?
AI enablers — chips, compute hardware, upstream suppliers — have doubled their forward EPS over the past two years; earnings momentum remains strong.
AI adopters — companies across industries using AI to cut costs and lift productivity — are projected to grow forward-12-month EPS by roughly 70% on a two-year cumulative basis. The earnings inflection is here.
High-AI-impact adopters are expected to expand EBIT margins by 460 basis points in 2025–2026, nearly twice the MSCI World average — yet the market has not fully priced in the long-run productivity dividend, leaving room for upward earnings revisions.
Are valuations still stretched?
High-AI-impact adopters now trade at a forward P/E of 18×, versus 22× for enablers. After the de-rating, the risk-reward profile has improved markedly.
This means → the adoption side is seeing an earnings inflection and cheaper valuations at the same time — what Morgan Stanley calls a value window.
What goes on each end of the barbell?
Bottleneck end (defend): prioritise power-related plays — energised-facility operators, energy storage, grid equipment, renewables. Within semis and infra, stock-pick around second-tier supply-chain bottlenecks beyond power.
New-allocation end (attack): add early-stage software enablers in order — infrastructure software → cybersecurity → select application software. Among AI adopters, screen on two metrics: the real-world impact of AI on the business and pricing power.
In plain terms = only companies that can keep the savings AI generates — rather than passing them on to customers — will convert productivity gains into actual profit. IT services are a late-cycle beneficiary and not yet a priority allocation.
How does the AI playbook differ across regions?
North America: AI dividends are diffusing from hardware into software and the real economy.
Asia-Pacific: returns still concentrate in the upstream supply chain, but more companies are disclosing AI-driven revenue in earnings reports.
Europe: AI opportunities cluster in traditional-industry application deployments.
This reflects different regions sitting at different stages of AI diffusion. Morgan Stanley stresses that the degree of real AI impact on a business is the single most important alpha variable — companies facing a core-business AI threat will significantly underperform peers absorbing only moderate disruption.
市场有风险，内容仅供研究参考，不构成投资建议。