2026-11-10 09:25

Lead: Morgan Stanley says AI sector has fallen out of buying range post-midterm election, capital expenditure remains unchanged—recommend buying three categories on dips.

The median semiconductor stock has declined by 53% from its peak, and the Philadelphia Semiconductor Index has dropped around 40% since mid-June. On November 8, Morgan Stanley released a 93-page global thematic strategy report titled directly:「Buy the Drop」. The report argues that market reaction to the U.S. midterm election results was overblown—AI capital expenditure overall has not been downgraded, only about 20% of U.S. projects (10% of global projects) have been postponed from 2028 to 2029.

Election Impacts Timing, Not Volume

In the 2026 midterm elections, Democrats secured control of both chambers of Congress and won gubernatorial races in Texas, Ohio, and Georgia. The incoming Texas Governor Gina Hinojosa proposed suspending new data center approvals, while Ohio’s incoming Governor Dr. Amy Acton advocated for conditional suspension. Market panic quickly spread across the entire AI supply chain. The report asserts that ongoing projects will not be disrupted, and 2027 capacity forecasts remain unchanged. The impact is concentrated in 2028: approximately 5GW of IT capacity (4% of U.S. total planning) from Texas and Ohio will be delayed until 2029. In other words, the election alters construction timing—not AI demand itself. The report maintains its forecast of $1.6 trillion/$1.8 trillion in capital expenditures for hyperscale cloud providers in 2027/2028.

Postponement Extends the AI Investment Cycle

The report runs the numbers: U.S. data center capacity additions in 2028 are revised downward from the prior forecast of 24.9GW to 19.9GW,with year-on-year growth shifting from positive to -4%. However, the delayed capacity hasn’t vanished—it is stacked into 2029, causing annual additions to surge from 22.7GW to 27.6GW, with year-on-year growth reaching 38%.

The report identifies three ancillary benefits of this “postponement rather than cancellation” dynamic: smoother rebalancing of data center supply chain dynamics, improved credit market outlook, and rising compute rental pricing. Given constrained power availability in the U.S., projects with grid access become scarcer, giving hyperscale cloud providers priority in resource allocation.

On a global scale, total capacity in 2028 is revised downward from 224GW to 219GW, while capacity beyond 2029 remains stable at 271GW. Power deficit expands from 38GW to 44GW, representing 38% of required electricity.

Morgan Stanley Breaks Down “Buy the Drop” into Three Baskets

The report framework is clear: three categories, each with distinct rationale.

First basket: Hyperscale cloud providers. The report favors companies with strong scale, capital access, and LLM technology iteration capabilities. Brian Nowak upgrades META, GOOGL, and AMZN to Overweight; Adam Wood upgrades MSFT to Overweight. Rationale: these firms have diversified geographic footprints, are not dependent on any single state, and have not adjusted their capital expenditure guidance.

Second basket: Select data center supply chain. The report recommends buying power and infrastructure companies that have already declined below intrinsic value during corrections, while cautioning against potential inventory buildup risks in the supply chain by 2028. High-conviction names include BE (Bloom Energy), SEI (Solaris), VST (Vistra), TLN (Talen Energy), and CEG (Constellation Energy). Even if data center growth falls short of expectations, the risk-reward profile remains favorable. For colocation providers, EQIX (Equinix) has limited exposure in Texas (only 9 operating facilities, accounting for ~3% of its global 280 facilities), potentially benefiting from higher pricing.

Third basket: Software enablers and AI adopters. The report expects value creation to shift from semiconductors and infrastructure toward software and application layers. Cybersecurity and infrastructure software are preferred entry points—upgrades to PANW (Palo Alto Networks) and CRWD (CrowdStrike). In SaaS, upgrades to SNOW (Snowflake), DDOG (Datadog), and MDB (MongoDB).

Semiconductor Declines Deep Enough—Report Says It’s Time to Pick Up

The report uses a chart to illustrate the point: as of November 6, the median semiconductor stock in Morgan Stanley’s coverage had declined 53% from its 52-week high, with an average decline of 48%. The report views this as a buying opportunity because the core pillar of AI spending (demand) remains intact, and token usage continues to grow exponentially.

Compute chips are the highest-conviction sub-sector in the report. NVDA and AVGO both trade at approximately 9x CY28 P/E. NVDA is the top pick, with three key reasons: broader customer base than AMD/ASIC, ability to grow revenue per watt even under power constraints (next-gen Feynman architecture expected to increase revenue per GW from Rubin’s ~$40 billion to over $500 billion), and leadership in tokens/watt metrics.

For memory, preference leans toward MU and SNDK. The report specifically highlights SanDisk: based on the closing price of $835 as of November 6, if buyback size reaches around 50% of revenue, over 50% of outstanding shares could be repurchased by end-2028.

AI Adoption Has Shifted from Experimentation to Measurable Returns

The report cites analysis from AlphaWise on roughly 18,000 earnings reports and meeting transcripts. Among S&P 500 companies, 44% mentioned at least one quantifiable AI benefit in the past 12 months—a rise from 32% a year ago. Companies citing cost savings or AI-driven revenue growth rose from 16% to 26%.

The financial sector leads the way: 67% of financial firms referenced tangible improvements from AI, up from 42% a year ago. The report advises broadening investment focus—from AI builders to AI adopters, including banks, insurers, and consumer-facing firms.

Risk Disclosure

This article is a compilation and interpretation by ChaoXiang Research of third-party brokerage research reports. The ratings, target prices, earnings forecasts, and related judgments cited herein represent the views of the analysts at the originating institution, reflect only their institutional stance, and do not constitute ChaoXiang Research’s opinion or any investment advice.

Markets involve risk—decisions must be independent. This article should not be used as the basis for any securities transaction.

Source: TechFlow Column