The U.S. midterm election results will likely reshape the pace of the AI buildout without fundamentally changing its ultimate size (see the 4 November 2026 article, “With Echoes of 1974, a Decisive Victory for the Democratic Party”). More permitting friction, higher costs, and longer development timelines may slow capex and therefore reduce near-term financing needs.
The potential exposure to such a slowdown is significant. Data compiled from Cleanview suggest that roughly 45% of planned U.S. data center capacity is located in states where tighter restrictions could emerge, with Texas alone representing roughly one-quarter.
That said, 45% exposure does not necessarily mean 45% is at risk. Projects may be delayed, relocated, or redesigned rather than canceled; some of today’s backlog would probably never have been built anyway. The ultimate impact will depend on the scope and timing of new restrictions and the permitting status of individual projects.
Perhaps more importantly, unless the underlying demand for compute changes materially, the long-term need for capacity – and the financing behind it – should still be substantial.
For now, however, we expect a longer, and less linear, AI buildout rather than a materially smaller one.
We don’t expect that path to be smooth. The past two weeks have seen plenty of noise across the AI ecosystem. The NVIDIA governance change, CoreWeave’s upside earnings surprise, and the macro implications of the midterms are among a confluence of catalysts driving substantial volatility in both equity and credit.
Once the dust settles, we expect the underlying differentiation to reassert itself: A slower AI buildout should favor the players best able to absorb delays, while others may struggle amid uncertainty.
Scarcity favors those that can afford to wait
With their generally strong balance sheets and varied earnings sources, diversified hyperscalers appear well-positioned to withstand a slower investment cycle.
Slower deployment extends the scarcity of existing compute, supporting utilization and pricing while reducing the risk of eventual overcapacity. Their balance sheets, diversified cash flows, and geographic flexibility also give these hyperscalers more options to redirect investment as permitting conditions change.
However, diversified hyperscalers are not completely immune to an AI slowdown. Physical delays do not necessarily delay financial commitments. GPU purchases, leases, and infrastructure obligations remain in place even if revenue gets pushed further into the future. And hardware bought for delayed sites could depreciate before it is fully monetized. The impact on free cash flow therefore depends on how much spending can be genuinely deferred versus how much is already committed.
This tension creates an interesting asymmetry between equity and credit markets. Equity tends to absorb both the upside from prolonged scarcity and the potential drag on investment returns. Credit can benefit from a less front-loaded AI financing calendar and therefore a slower pace of issuance. Even if cumulative financing needs remain large, stretching them over a longer period reduces the near-term technical pressure from supply.
The same logic is likely to favor existing hyperscaler-tenanted data center debt, particularly issues with strong structural protections for bondholders. If slower construction reduces data center issuance more than senior unsecured issuance, the ensuing technical tailwind should support spread compression between hyperscaler-tenanted data center bonds and the senior debt of their tenants (see Figure 1). For bondholders, the strongest structures are those where contractual protections keep the tenant economically responsible even when construction is delayed.
Scarcity offers breathing room, but it is no panacea
The calculus is very different for more leveraged infrastructure providers.
On one hand, existing powered capacity benefits from scarcity. On the other hand, these businesses also have less balance sheet flexibility and rely more heavily on continued access to financing. Leases, GPU commitments, and interest expenses may continue while construction and revenues are delayed; a slower issuance calendar provides little comfort if existing capital remains tied up for longer than planned.
This creates a tension between scarcity value and valuation. Scarcity can support utilization, rental rates, and near-term cash flows. But investors still need to value cash flows beyond the initial contract, where economics depend on future demand, recontracting rates, GPU residual values, and access to financing.
A slower buildout can allow older GPU capacity to be renewed at more attractive prices, but it raises a key question for equity and credit investors: What will that capacity be worth several years from now as customer requirements and technology evolve?
Therefore, for leveraged business models, the risk is that the scarcity premium may be more than offset by a valuation discount for recontracting risk and a longer, less certain monetization timeline.
Slower does not mean smaller
Before the election, consensus forecasts pointed toward another extraordinary period of AI investment, with hyperscaler capex estimates of roughly $1.2 trillion in both 2027 and 2028 and investment running ahead of operating cash flows (see Figures 2 and 3). Those forecasts may now be revised lower or, more likely, pushed further into the future.
For credit portfolios, a slower AI buildout may reduce the flow of new issuance, but it does not make the concentration challenge disappear, as we discussed in the 3 November 2026 article, “The Credit Market Lens: Where to Seek Diversification From the AI Investment Theme.” The stock of AI-related exposure is already substantial, and the eventual financing requirements could remain large even if they arrive over a longer horizon.
The U.S. election outcome may slow the AI buildout, but it’s unlikely to stop it. For investors, the key post-election questions are who earns the scarcity premium in the interim, who bears the cost of delay, and who has the balance sheet capacity to bridge the two.
Michael Puempel and Gabriel Cazaubieilh contributed to this report.