Thoughts on the Market
2026-09-24 · Hosted by Mike Wilson · Morgan Stanley
Executive Summary
Morgan Stanley's Ariana Salvatore argues that a growing focus on AI safety is likely to be a tailwind for compute spending and capital expenditure (CapEx, spending on data centers and infrastructure), not the brake many investors fear. Across five potential regulatory scenarios, Morgan Stanley expects labs to spend more, not less, as they build out safety monitoring infrastructure alongside rapidly advancing model capabilities.
Key Stories & Changes
1. AI Safety Focus Seen as a Tailwind, Not a Brake, on CapEx
Morgan Stanley's base case: increased safety monitoring infrastructure requirements will accelerate, not reduce, compute spend, especially as LLM capabilities increase at a nonlinear rate
Team modeled five potential regulatory states of the world, ranging from industry self-policing to heavier government intervention — all pointing toward higher, not lower, CapEx
Framework rests on two variables: incident salience (how prominent/damaging a triggering event is) and instrument availability (whether government already has a usable regulatory tool)
2. Three Obstacles to Comprehensive AI Regulation
Politics: President Trump has publicly opposed large-scale AI regulation
Procedure: No clear legislative vehicle currently exists for a comprehensive bill to attach to
Precedent: Historically, major regulation (e.g., the CARES Act post-pandemic, ARRA post-2008) follows a high-salience triggering event, not gradual concern-building — and no such event has yet occurred for AI
3. No Sweeping Open-Weight Model Regulation Expected
Morgan Stanley does not expect a broad crackdown on open-weight models, for three reasons:
Congressional proposals like the Kill Switch Act (mandating companies retain ability to shut down models on short notice) are seen as less likely to pass than administration-led incremental guardrails
4. Data Center Pushback Unlikely to Meaningfully Slow Buildout
Current opposition to data centers is driven mainly by environmental concerns and affordability, not AI safety risk specifically
Pushback is highly idiosyncratic by state and locality; hyperscalers have shown willingness to make targeted concessions (e.g., Google publishing water usage data, Meta announcing a local community fund)
Morgan Stanley frames the buildout as increasingly "conditional" — contingent on concessions and potentially costlier in some regions — but not derailed in aggregate
5. Midterms Seen as a Sentiment Driver, Not the Main Policy Determinant
Salvatore argues the 2026 midterms won't be the primary driver of the AI regulatory path because President Trump remains in office through 2029 regardless of outcome, and any bill would still require cross-party collaboration with the president
The more important variable remains whether a high-salience triggering event occurs, which could catalyze regulation "regardless of [congressional] government configuration"
6. Investment Guidance: Own Compute Bottlenecks, Cybersecurity, and AI Adopters
Morgan Stanley reiterates three thematic recommendations:
Separately, the firm recommends avoiding negative exposure to rising US-China technology transfer restrictions, expecting continued bifurcation of the global AI market
Trends Identified
1. Safety Spend and Capacity Spend Are Becoming Complementary, Not Competing
Morgan Stanley's core thesis reframes AI safety compliance as additive to, rather than a drag on, the AI capital spending cycle — as labs build monitoring and safety infrastructure alongside frontier model development, both categories of spend rise together rather than trading off.
2. Regulatory Catalysts Require a Triggering Event, Not Gradual Pressure
The historical pattern cited (CARES Act, ARRA) suggests markets should watch for a specific high-salience AI incident as the key swing factor for regulation, rather than assuming steady political pressure alone will produce comprehensive legislation.
3. Fragmented, Local-Level Regulation Is the More Likely Near-Term Path
Rather than sweeping federal action, AI-related friction is more likely to manifest through idiosyncratic, localized data center pushback (water usage, community impact), which hyperscalers can generally absorb through targeted concessions. ---
Sentiment Analysis
Overall Market Sentiment: Constructive
Morgan Stanley's research team is explicitly bullish on the durability of the AI CapEx cycle, framing safety concerns as a net positive catalyst for compute demand rather than a regulatory risk to the investment thesis.
Risk Factors Highlighted
High-salience triggering event: A significant AI safety incident could rapidly catalyze regulation regardless of current political gridlock.
Kill Switch Act and similar congressional proposals: Though seen as less likely to pass, such bills represent tail risk to the CapEx thesis if incident salience rises.
Localized data center opposition: Idiosyncratic state/local pushback on environmental and affordability grounds could raise costs or delay specific projects even if it doesn't derail the aggregate buildout.
US-China technology transfer restrictions: Increasing bifurcation of the global AI market creates risk for companies with exposure to cross-border technology transfer.
Enforcement gaps in open-weight models: The inability to effectively control distribution and downstream use of published model weights remains an unresolved policy and safety challenge.
This episode was covered in today's [The Market Signal — 2026-09-24](https://marketsignal.beehiiv.com/p/the-market-signal-2026-09-24), a cross-source synthesis of multiple podcast reports.