Thoughts on the Market
2026-11-16 · Hosted by Mike Wilson · Morgan Stanley
Executive Summary
Morgan Stanley's Tom Wigg moderated a discussion between Stephen Byrd (Global Head of Thematic Research) and Joe Moore (US Semiconductors Analyst) on whether AI chip demand can outrun a worsening power shortfall. Morgan Stanley had already estimated a 30-gigawatt shortfall between now and 2028, and Byrd said that estimate has since been revised into the mid-30s gigawatts following midterm election outcomes in Texas and Ohio, where new governors are expected to require longer permitting processes for data centers. Byrd characterized this as a delay, not a cancellation — the buildout likely shifts weight into 2029, which he called a "relief valve" that elongates the overall AI growth cycle.
Key Stories & Changes
1. US Data Center Power Shortfall Revised Higher
Morgan Stanley's prior estimate: 30-gigawatt shortfall between now and 2028
Revised estimate: shortfall now in the mid-30s gigawatts, driven by midterm election outcomes in Texas and Ohio
Byrd: new governors are expected to require longer permitting/community-engagement processes for data center developers, causing a delay, not a cancellation
The delayed capacity is expected to show up in a much larger 2029 deployment, acting as a "relief valve" that extends rather than shortens the overall AI buildout cycle
2. Potential Offsets to the Power Shortfall
Off-grid generation: a viable category, but engineering, labor, and permitting timelines limit near-term impact; Virginia has effectively outlawed on-site gas generation over emissions concerns
International expansion: Malaysia (Google recently signed a deal there), India, Thailand, Japan, and China cited as incremental options; but other countries are adding restrictions too — Norway has a data center moratorium, the UK saw a major project terminated over grid access delays, and Australia faces its own issues
Grid optimization: Byrd is explicitly not bullish here — despite the US grid being utilized only ~50% of the time on average, utilities plan for worst-case scenarios (e.g., a six-day winter storm) and won't flex capacity for data centers under that conservative mindset
Community-oriented development deals: Louisiana cited as a model, where developers are funding new hospitals and pay raises for teachers, police, and fire services in exchange for local support
3. Semiconductor Demand Remains Tight Despite Power Concerns
Joe Moore: for 2027, chip supply is "not just staying tight. It's getting tighter" — multiple hyperscaler CEOs cite memory, not land/power, as their biggest near-term bottleneck
Moore's framing: ubiquitous compute demand means capacity freed up in one location (e.g., a delayed Texas or Ohio project) tends to find a buyer elsewhere
Memory stocks are trading at roughly three times earnings, down 60-70% from highs, which Moore called an attractive entry point if sustainability holds
Moore: "the dissonance between Silicon Valley optimism and Wall Street pessimism has never been this wide"
4. Political Dynamics Around Data Center Buildout
Byrd: the midterm election results reflected genuine local concern about poorly planned data center development, which he called "warranted in some cases"
He expects "smart approaches" — community investment, infrastructure give-backs — to unlock continued buildout even in states with newly skeptical governors
Tom Wigg noted investor debate over whether blue-state governors will ultimately prioritize economic activity over campaign rhetoric once in office
Trends Identified
1. Power Constraints Are Reshaping Timing, Not the AI Investment Thesis
Both guests agreed the core AI infrastructure demand story remains intact; the primary effect of the power shortfall is to redistribute buildout timing further into 2028-2029 rather than reduce total expected capacity. This reframes near-term bearishness on semis as a duration question rather than a demand destruction question.
2. Memory, Not Land or Power, Is the Binding Constraint Today
Despite headline concerns about power and permitting, Joe Moore's channel checks show hyperscalers currently naming memory shortages as their most pressing bottleneck. This suggests near-term semiconductor stock performance may be more sensitive to memory supply dynamics than to the power narrative dominating headlines.
3. Political Risk to Data Center Buildout Is Becoming State-Specific and Negotiable
Rather than a uniform national headwind, the political risk to AI infrastructure is emerging state-by-state, with outcomes shaped by whether developers offer local community benefits. This creates a more fragmented, negotiable risk landscape than a simple national regulatory crackdown would imply.
4. Valuation Divergence Between Public Markets and Private AI Optimism
The widening gap between Silicon Valley's continued confidence and Wall Street's skepticism — reflected in heavily discounted memory valuations — suggests public equity markets may be underpricing the duration of the AI capex cycle relative to insider sentiment. ---
Sentiment Analysis
Overall Market Sentiment: Constructively Bullish on Fundamentals, Cautious on Near-Term Politics
Both guests expressed confidence in underlying AI chip demand durability through 2028 even as they acknowledged genuine, election-driven delays to the physical infrastructure buildout.
Risk Factors Highlighted
Widening power shortfall: The US data center power gap has been revised from 30 to the mid-30s gigawatts following election-driven permitting delays in Texas and Ohio.
State-level political risk to data center siting: New governors may impose longer, more restrictive permitting processes that delay (though not cancel) capacity buildout.
Grid optimization is not a viable near-term fix: Utilities' conservative planning for worst-case demand scenarios limits any ability to flexibly serve data center loads from existing slack capacity.
International siting risk is also rising: Moratoriums or project cancellations in Norway, Australia, and the UK show that offshoring data center capacity is not a frictionless offset.
Memory and component shortages persist: Even if land and power constraints ease, binding shortages in memory could still cap near-term AI infrastructure deployment.
Duration-trade fatigue: Moore warns that continually having to "prove" each future year's demand (2027, then 2028) creates recurring risk to semiconductor stock sentiment even if the long-term thesis remains intact.
Valuation/sentiment dissonance: A persistent gap between private AI-industry optimism and public market skepticism could resolve in either direction, creating volatility risk for memory and semiconductor equities.
This episode was covered in today's [The Market Signal — 2026-11-16](https://marketsignal.beehiiv.com/p/the-market-signal-2026-11-16), a cross-source synthesis of multiple podcast reports.