Data center capex story is still intact, despite pushback, says Morgan Stanley's Salvatore

Market impact: Mixed for hyperscalers, constructive for the AI infrastructure complex.

The key market implication is that local and political resistance is being interpreted as a supply-side constraint rather than a collapse in AI demand. Morgan Stanley’s broader thesis is that power availability, permitting, land, and grid access are becoming the limiting factors; this can delay individual projects while encouraging hyperscalers to secure sites, electricity, and equipment earlier. The firm has estimated total 2026 AI capital expenditure, including smaller cloud providers, at roughly $870 billion, with risks skewed higher if companies accelerate capacity reservations.

For Microsoft, Amazon, Alphabet, and Meta, this is initially mixed:

• Bullish: Scarce power and permitted capacity can reinforce the strategic value of existing sites and raise barriers to entry, supporting cloud pricing and long-term AI infrastructure demand.
• Bearish: Delays increase construction costs, extend the period before capacity generates revenue, and pressure free cash flow. Investors may become less willing to reward higher capex unless cloud revenue and AI monetization accelerate proportionally. Recent reporting indicates that hyperscaler spending is already producing greater scrutiny over margins, debt issuance, and the gap between capex growth and revenue growth.

The more direct beneficiaries are likely to be power, electrical-equipment, cooling, construction, and data-center infrastructure suppliers. If projects are relocated rather than canceled, demand may rotate toward companies exposed to transmission, substations, backup generation, natural gas, nuclear power, liquid cooling, and facility construction. However, project delays can create uneven order timing and raise execution risk, so the news is more supportive of the sector’s long-cycle demand outlook than necessarily of near-term quarterly results.

For utilities and power generators, the theme can support higher load-growth expectations and infrastructure investment, but the effect is two-sided. Utilities may benefit from large-load contracts and rate-base expansion, while political scrutiny raises the risk of moratoriums, cost-sharing requirements, or restrictions designed to prevent households from absorbing data-center-related grid costs. The market will likely distinguish between companies with secured generation and transmission capacity and those dependent on uncertain interconnections.

The main bearish interpretation is that resistance becomes sufficiently widespread to cause permanent demand destruction, not merely geographic relocation. Data cited in related coverage shows at least $156 billion of projects were canceled or delayed amid opposition in 2025, while construction capacity in major U.S. markets declined in the second half of that year because of permitting, zoning, and power-procurement hurdles. At the same time, equipment purchases and development in more accommodating regions continued, supporting the argument that the buildout is being reshaped rather than abandoned.

Likely time horizon:

• Short term: Higher volatility in megacap technology and AI-linked equities as investors reassess capex timing and returns.
• Medium term: Relative support for power and infrastructure suppliers, with project mix shifting toward regions offering faster permitting and reliable electricity.
• Longer term: The decisive issue remains whether AI services produce sufficient revenue and utilization to justify the extraordinary capital intensity.

Traders should monitor hyperscaler capex guidance, AI-cloud revenue growth, free-cash-flow conversion, data-center lease commitments, utility interconnection queues, state tax incentives, permitting decisions, and evidence of power-cost pass-through to customers. The most important confirmation would be continued increases in AI demand and contracted capacity despite slower project approvals; the clearest invalidation would be broad capex reductions accompanied by weaker cloud growth and falling equipment orders.