2026 midterms put AI data centre power and permits in focus

The November 3 vote can change congressional control, but Washington may not control the next AI buildout decision. Power, permits and energy costs are moving to states and cities.

The November 3 US midterms put congressional control in play. The consequential distinction is between policies Washington can change and infrastructure decisions it cannot directly deliver.

All 435 House seats and 35 Senate seats were contested. The Associated Press election report framed the vote as a referendum on President Donald Trump's second term. That framing leaves the eventual balance of power unresolved.

CTOL's thesis is that congressional outcomes can influence financing conditions, while state and local decisions may determine which AI data centres actually enter service. Capital commitments are not operational capacity.

What changes in Washington

Congressional control changes committee leadership, oversight, budget negotiations and legislative prospects. It does not automatically reverse tariffs, tax rules or agency decisions.

Morgan Stanley's election analysis examines congressional control and market implications. Its separate midterm policy analysis argues that executive and agency action can sustain parts of the administration's agenda under divided government.

How fiscal policy affects markets matters more than the party label.

A Congress willing to expand tax relief or spending could increase expected borrowing, placing upward pressure on longer-term Treasury yields. Higher yields could increase infrastructure financing costs and reduce the present value of future earnings.

Gridlock could instead constrain new fiscal commitments while making funding deadlines and debt-ceiling negotiations more disruptive. Lower expected spending would not necessarily mean lower market risk.

Unified control could accelerate legislation while increasing exposure to deficit concerns. Divided control could constrain legislation without stopping executive trade or regulatory action.

The signal to watch is whether budget proposals, enacted measures and Treasury yields confirm the fiscal expectations investors price after the election.

AI regulation is a federal and local story

The Axios election analysis describes AI as an increasingly prominent political issue.

But AI infrastructure faces two policy systems.

Federal decisions can affect trade costs, tax incentives, regulatory obligations and infrastructure funding. State governments, utility commissions and local authorities can influence electricity prices, grid connections, zoning and permits.

Morgan Stanley describes a conditional data-centre buildout, with investment dependent partly on energy costs, infrastructure funding and community impact.

The process is straightforward. A developer may secure customers, chips and financing yet miss its operating schedule without power access or permits.

Delays can postpone equipment deployment and revenue while financing costs continue. Capital may shift toward locations with available power and faster approvals.

That creates a potential divergence between AI demand, announced investment and productive capacity.

The strongest counterargument is federal intervention. Binding national permitting rules, infrastructure funding or effective regulatory pre-emption could accelerate approvals and weaken local constraints.

CTOL therefore treats local dominance as a testable expectation, not an established fact.

Markets have a historical pattern, not a guarantee

Morgan Stanley reports that the S&P 500 historically gained an average 23% in the year after midterms when a Republican president faced split congressional control. The comparable figure was 15% when a Republican president faced a Democratic Congress.

The eight-percentage-point difference is descriptive, not a measured election effect.

These averages do not isolate monetary policy, earnings, starting valuations or external shocks. The supplied material provides neither sample sizes nor statistical significance. They cannot support a reliable 2027 return forecast.

The useful comparison for AI infrastructure is instead between capital-spending ambitions and projects obtaining usable power and approval.

Morgan Stanley estimates that hyperscaler capital spending could reach $1.5 trillion in 2027 and that power shortfalls could persist through 2028.

Both are projections, not realised outcomes. Together, they suggest a possible mismatch between investment ambitions and the electricity infrastructure required to support them.

That mismatch could separate companies benefiting from project announcements from businesses whose revenues depend on completed, energised facilities.

Federal incentives may improve project economics immediately. They cannot alone guarantee grid capacity, local approval or community acceptance.

Investment follows two timelines: policy announcements can change valuations quickly, while physical approvals, equipment deployment and commissioning determine when capacity becomes productive.

Possible outcomes

The base case is fragmented authority. During the first three months, congressional committees establish agendas while state and local processes continue.

By twelve months, permits and power access remain more direct constraints on individual projects than new federal AI legislation. Over 36 months, capacity growth depends on whether grid and permitting bottlenecks ease.

Evidence from committee calendars, local approvals, utility interconnection milestones and commissioned capacity would strengthen this case. Faster federal-driven approvals would weaken it.

An alternative is effective federal intervention. Unified control or cross-party agreement could accelerate tax, trade, infrastructure or permitting measures.

Within three months, concrete bill text and legislative timetables would strengthen this scenario. By twelve months, enacted measures would need to change approvals or project economics. By 36 months, faster operational capacity would provide stronger confirmation.

Legislation without measurable implementation would not establish success.

The downside is fiscal friction meets physical delays. Narrow congressional margins could intensify funding disputes while power constraints persist independently.

Within three months, repeated budget confrontations would be an early warning. By twelve months, higher financing costs alongside delayed permits or interconnections would pressure project returns. Over 36 months, persistent commissioning delays would challenge expectations that announced spending converts efficiently into capacity.

The downside need not be weaker AI demand. It could be a slower, more expensive conversion of demand into operating assets.

Evidence to watch

CTOL expects state and local decisions to determine more AI infrastructure capacity than a new federal AI law during the twelve months following November 3, 2026.

The relevant records are congressional committee calendars, enacted legislation, state and county permits, utility interconnection milestones and evidence of completed, energised facilities.

The central call fails if Congress enacts a binding national permitting or data-centre rule that materially changes project approvals by November 3, 2027.

It weakens if federal measures demonstrably accelerate operational capacity more than local approvals and power availability explain. Comparing those effects requires project-level evidence, not policy announcements alone.

The election can reprice policy risk. The decisive investment question is whether changed expectations produce approved projects, usable electricity connections and operating data centres.