We didn't see the zoning board as a geopolitical risk. But in 2025, the US midterm elections are rewriting the map for AI infrastructure, and the ripple effects are hitting the crypto side of the capital stack. The next 18 months will determine whether the $2000 billion in planned capital expenditure from Microsoft, Google, Amazon, and Meta actually materializes — or gets stuck in a political quagmire.
This isn't about campaign rhetoric. It's about county commissioners, environmental impact statements, and power purchase agreements. The AI infrastructure trade is now a political trade. And the market is underpricing the friction.
Context: The Physical Bottleneck
AI infrastructure has moved from the cloud to the ground. Every large language model needs a data center. Every data center needs land, power, and permits. The scale is staggering: a single hyperscale facility can consume 100+ megawatts — equivalent to tens of thousands of homes. Microsoft, Google, Amazon, and Meta are projected to spend over $2 trillion combined on capex through 2026, with the bulk allocated to AI-optimized data centers.
But these centers are not being built in a vacuum. They are sited in specific counties, subject to local zoning laws, tax incentives, and community sentiment. The US midterm elections on November 5, 2025, will determine control of the House, Senate, and numerous state legislatures. That shift will directly impact the regulatory environment for data center construction.
I've tracked this capital flow since 2021. The friction is real. In 2022, I saw a planned 300MW facility in Northern Virginia get delayed by 18 months due to local opposition over water usage. In 2024, a project in Ireland was capped by the grid operator because data centers were consuming 18% of the national electricity. These are not outliers. They are signals of a structural bottleneck.
Core: The Mechanical Friction of Political Risk
Political risk in AI infrastructure is not about policy changes in Washington. It's about the granular mechanics of site selection and permitting. Here's the data:
- Energy cost volatility: Data centers are power-intensive. A 10% increase in electricity costs due to local utility rate hikes or carbon taxes can reduce a facility's IRR by 200-300 basis points. Midterm elections often bring energy policy shifts — candidates in energy-rich states (Texas, Pennsylvania) are already campaigning on "AI power first" vs. "community protection."
- Permitting delays: The average build time for a hyperscale data center has stretched from 24 months to 36 months over the past two years, driven by environmental reviews and community hearings. A change in local government can add another 6-12 months as new officials renegotiate tax abatements.
- Community opposition: The backlash is real. In Chile, a $1 billion Google data center faced protests over water consumption in a drought-prone region. In Spain, Amazon's planned facility was delayed by a coalition of farmers and environmentalists. These are not NIMBY-ism — they are legitimate resource allocation conflicts. The midterm elections will amplify these voices as candidates court local votes by opposing "big tech data centers."
From a capital markets perspective, this friction is a hidden cost. The yield on AI infrastructure investments — whether through direct equity, REITs, or project finance — is being compressed by rising political risk premiums. I've audited multiple data center REITs. The ones with a concentrated geographic footprint (e.g., all in Virginia or California) have higher sensitivity to local political shifts. The diversified ones, with exposure to states like Texas, Arizona, and Ohio, are better positioned.
But here's the crux: the market is still pricing AI infrastructure as a pure growth story. It's not. It's a regulated utility with a growth veneer. The midterm elections will force a repricing.
Contrarian: The Decoupling Thesis
The conventional wisdom is that AI infrastructure is a "must-build" — that the demand for compute is so insatiable that political friction will be overcome. I disagree. The decoupling thesis is this: the AI narrative and the physical infrastructure buildout will diverge by 2026.
Why? Because the technology is moving faster than the politics. AI models are becoming more efficient, with smaller models (e.g., Mistral, Phi-3) achieving competitive performance with less compute. Simultaneously, alternatives like edge computing and decentralized compute networks (think Filecoin, Akash) are emerging as lower-friction options. These don't require massive land acquisitions or power purchase agreements. They run on existing hardware.
So the political risk isn't just a delay. It's a catalyst for a structural shift away from hyperscale centralized data centers towards distributed, modular infrastructure. This is good for crypto-native compute projects, but bad for the traditional AI infrastructure trade that has been the darling of institutional investors.
Moreover, the midterm elections could produce a bifurcated regulatory landscape. A Republican-controlled House might push for faster permitting and tax incentives for data centers in red states, while Democratic-controlled states (California, New York) impose stricter environmental and labor standards. Capital will flow to the path of least resistance. The result: a geographic concentration of AI infrastructure in politically friendly states, but a shortage in high-demand, high-cost states. This will create pricing inefficiencies that sophisticated traders can exploit.
Takeaway: Watch the County Clerk, Not the Conference Keynote
Yields don't flow evenly. They flow where the friction is lowest. The midterm elections are adding friction to the AI infrastructure trade. The next 12 months will reveal which projects are real and which are paper. I'm watching local zoning board meetings, not GPU benchmarks. The real bottleneck is in the county courthouse.
For crypto investors, this is a macro signal. If AI infrastructure becomes politically constrained, the narrative shift towards decentralized compute will accelerate. If it doesn't, the centralized hyperscalers will dominate. Either way, the midterm elections are the event that will break the tie.
We didn't see this coming. But the data doesn't lie. The yield on AI infrastructure is about to get a haircut.
