The Democratic Sweep Means Increased Risk for GPU Megadeals

Tuesday's election isn’t going to stop the AI buildout, but it will shift who gets financed, and on what terms. Democrats won control of both the House and Senate, and in the states where data centers are actually being built, voters have now elected officials who promised to make data centers pay the full cost of the new power plants and transmission lines.

For GPU lenders, the lesson is simple: the political risk of very large projects just went up. The best risk-adjusted opportunity now sits in the Compute Middle Market, GPU deals between $25 million and $500 million.

Data centers were on the ballot in a way they have never been before. Electricity bills became a central midterm issue, with bills in Ohio, Michigan and Pennsylvania rising fastest, and those states are also data center hot spots. In Ohio, Sherrod Brown campaigned as a data center skeptic and unseated Jon Husted.

The Blue Wave Pressure

Rather than starting in Washington, the real decision making is in statehouses and utility commissions, where siting, cost allocation and interconnection get decided.

Texas will be the bellwether for the rest of the country, as newly elected Democrat Governor Gina Hinojosa, who has called for an immediate moratorium on new data centers, defeated incumbent Greg Abbott. Texas might be the night's biggest worry for tech investors, because it is expected to supply roughly a quarter to a third of US data center capacity. Hinojosa has also proposed making data centers pay for transmission, screening projects for grid impact, ending tax breaks and giving communities more say.

In Georgia, Democrats took control of the Public Service Commission, which regulates Georgia Power, for the first time in decades. The new majority has pledged to hold data centers accountable for their impact on bills. Even where Republicans held on, as in Louisiana's PSC races, the winning candidate framed support for data centers as conditional on not raising rates.

Wedbush's Dan Ives now estimates that 10% of planned US data centers won't get built as a result of the election.

What It Means for Underwriting

For lenders, this changes how GPU deals get priced. Three risks are now becoming more obvious.

Permitting risk: Local opposition was already rising before Tuesday. Data Center Watch counted at least 75 projects worth about $130 billion blocked or delayed in Q1 2026 alone.

Oracle's Project Jupiter in New Mexico is a gigawatt-scale buildout that will be under the microscope. The roughly 2.5 GW Stargate campus has faced resident opposition, lawsuits and two state rejections of the gas pipeline meant to power it. In September, Oracle sent a force majeure notice to the developer, a Blue Owl unit, seeking to delay payments if the site misses its 2028 target, and the roughly $18 billion of debt tied to it was reported to be trading at stressed levels. Oracle says the project remains on schedule, but lenders took notice. Elected officials now have proof that opposing data centers wins votes, and CNBC quoted one analyst warning that local officials will see political upside in blocking or delaying them.

Power cost risk: When regulators shift transmission and grid upgrade costs onto large loads, a project's operating costs change after the loan closes. Fitch had already downgraded the North American utility sector's outlook in June, citing political pressure to limit bills and utility profits.

Timeline risk: Interconnection in the largest markets already takes four to seven years. Stricter grid screening adds more time, and GPUs depreciate while they wait.

All three risks grow with project size as a gigawatt campus needs new generation, new transmission and a public vote. This can be a single point of failure for a large GPU project, with billions of dollars of exposure riding on one county board and one utility commission.

Why Capital Moves to the Middle Market

The Compute Middle Market is built differently. A $25 million to $500 million GPU deal typically goes into an existing powered shell, a colocation facility or an already-interconnected site. It doesn’t require a new campus that a county can vote down, or a multi-year wait for a new substation. The GPUs can be generating revenue within months, not years.

Those deals also diversify well with exposure spread across operators, regions, utilities and end customers, instead of concentrating in a handful of megaprojects tied to a handful of tenants. A pool of performing middle-market GPU loans can look more like the auto and equipment ABS markets institutions already buy than a single-asset bet on one campus.

Hyperscalers can fund gigawatt projects from their own balance sheets, but the gap is and will continue to be financing. Middle-market operators can't, and many banks still lack the data and structures to underwrite GPUs as collateral at scale. That leaves an opening for private credit funds and multi-asset managers willing to build the expertise now.

The election won’t end AI's demand for compute, but it has made the largest projects slower, costlier and more political. Capital that wants to stay exposed to AI growth without taking on that political risk should move toward the middle market.