Midterms Could Delay the AI Buildout, Not Derail It

Ariana Salvatore · Morgan Stanley

About 5GW of US data center projects slip from 2028 into 2029; capacity flat in 2028, up ~40% in 2029.

Morgan Stanley colleague Cameron McVeigh and team estimate that approximately five gigawatts of US data center projects, roughly 28% of the 2028 pipeline, will shift into 2029. 2027 construction is expected to be unaffected because projects are generally already permitted or underway. As a result, US data center capacity additions are projected to be roughly flat in 2028 before growing nearly 40% in 2029.

“As a result, [U.S.] Data center capacity additions are projected to be roughly flat in 2028 before growing nearly 40% in 2029.”at 2:01

Ariana Salvatore · Morgan Stanley

The largest, best-capitalized hyperscalers are best placed to absorb delays and higher costs.

The largest and best-capitalized hyperscalers are particularly well positioned in the power-constrained build-out. They have line of sight to 25 to 50% returns on investment, flexibility to redirect investment, the ability to absorb higher costs, and the means to improve computing efficiency.

“The largest and best capitalized hyperscalers are particularly well positioned. These are the companies that have a line of sight to 25 to 50% return on investment.”at 2:57

Ariana Salvatore · Morgan Stanley

A slower 2028 build-out risks weaker equipment orders, inventory build and supplier margin pressure.

A slower build-out in 2028 could mean weaker equipment orders, inventory accumulation and pressure on supplier margins. The speaker notes this could happen even if underlying AI demand remains strong, so the risk comes from the timing of data center construction rather than from demand.

“A slower build out in 2028 could mean weaker equipment orders, inventory accumulation, and pressure on supplier margins, even if the underlying AI demand remains strong.”at 3:20

Ariana Salvatore · Morgan Stanley

New data center construction financing faces tougher scrutiny; existing cash-flowing facilities gain lease pricing.

The team expects substantial AI-related borrowing in 2027 led by hyperscalers, but financing will become more differentiated. New construction financing will face greater scrutiny around permitting, power availability and execution. Existing data centers with stable cash flows could benefit from increased scarcity and stronger lease pricing.

“But new construction financing will face greater scrutiny around permitting, power availability, and execution. By contrast, existing data centers with stable cash flows could benefit from increased scarcity and stronger lease pricing.”at 3:45

Ariana Salvatore · Morgan Stanley

Midterms will shift where and when data centers get built and their cost, not whether they get built.

Morgan Stanley's view is that the midterms are more likely to change where data centers are built, when they come online, and what they cost, rather than whether they get built at all. Opposition over electricity bills, water use and economic benefits is treated as a longstanding risk that the election results reinforce. The team sees the friction as largely state-by-state and project-by-project. The base case is a delay rather than a fundamental cut in investment, and the speaker argues the constraints could actually extend the duration of the AI investment cycle rather than derail it.

“We think that the midterms were actually more likely to change where data centers get built, when they come online, and what cost is associated with doing so, rather than whether they get built at all.”at 0:34

Ariana Salvatore · Morgan Stanley

Data center opposition is state-by-state; Texas's proposed approval pause is the most consequential result.

Texas was the most consequential election result, with Governor-elect Hinojosa proposing a pause on new data center approvals until stronger protections are in place. In Ohio, Governor-elect Acton has proposed a conditional moratorium. Georgia's approach is more incremental, with greater scrutiny of construction approvals and of how electricity infrastructure costs are allocated. Opposition is increasingly bipartisan, but the friction remains primarily local.

“Texas, we think, was the most consequential result. Governor elect [Hinojosa] has proposed pausing new data center approvals until stronger protections are in place.”at 0:55

Ariana Salvatore · Morgan Stanley

Federal government stays accommodative on AI infrastructure given its national security importance and likely veto.

At the federal level, the government is expected to keep an accommodative posture toward data centers. The reasons are the importance of AI as national security infrastructure and the likelihood of a presidential veto on restrictive legislation.

“At the federal level, we think the government retains an accommodative posture, just given the importance of AI as national security infrastructure and the likelihood of a presidential veto on restrictive legislation.”at 1:29

Ariana Salvatore · Morgan Stanley

US data centers face a 44GW power shortfall through 2028, shifting the AI bottleneck from chips to power.

Stephen Byrd and the Powering AI team estimate a 44 gigawatt shortfall in US data center power through 2028, representing roughly 38% of projected demand. The speaker concludes that the critical bottleneck is increasingly shifting from computing equipment to power for data center facilities. This is one of the three bottlenecks (people, power, politics) the team has highlighted.

“Stephen Byrd and our Powering AI team now estimate a 44 gigawatt shortfall in US data center power through 2028, representing roughly 38% of projected demand.”at 2:37