A crypto publication covered a political rally last week. That should stop you. Crypto Briefing โ a venue whose readers are mostly miners, energy traders, and token speculators โ spent column inches on a Trump appearance in Ohio, defending AI data centers against bipartisan criticism. No token. No protocol. No chart. Just electricity, land, and a fight over who pays the bill. I read it three times hunting for the crypto angle and found it in the negative space. The reason a crypto outlet covers AI infrastructure is that both industries now bid for the same physical asset: the interconnection queue. That is the signal. The rally is noise. The grid is the story.
Here is what the report actually gave me. No timestamp. No named critics. No poll numbers, no electricity price data, no investment figures, no direct quotes. "Bipartisan criticism" as a phrase with nothing behind it. A source and subject mismatch โ a crypto outlet reporting political infrastructure news. I have to treat every specific claim as unverified. What survives that filter is the backdrop, and the backdrop is documented.
The United States entered 2024 and 2025 in a hyperscale buildout. Microsoft, Google, Amazon, Meta, and OpenAI's Stargate program announced capital expenditure in the hundreds of billions. Siting concentrated in power-rich, tax-friendly states: Virginia, Texas, and Ohio. Ohio sits inside PJM Interconnection, the largest regional grid operator in the country. PJM has repeatedly warned about capacity tightness. Residential electricity bills have become a political object in exactly the places where data centers land.
Trump's position is consistent with his "energy dominance" framing: AI leadership requires power, and power requires permits. But his populist base carries a triple grievance โ Big Tech, rising utility bills, and subsidies. That is the soil the criticism grows in. The rally is a public attempt to hold the coalition together. Whether it worked is unknowable from the source I was given. The structural tension is not.
Start with the economics, because the politics are downstream of them. A hyperscale data center signs long-term power purchase agreements, typically ten to fifteen years. Electricity is 20 to 40 percent of operating cost depending on PUE โ power usage effectiveness โ and the local tariff. A facility drawing 100 megawatts at a $60 per megawatt-hour contract spends roughly $52 million a year on power alone before a single rack is provisioned. Push the tariff up 15 percent and you erase tens of millions in annual margin. Math doesn't negotiate. No political endorsement changes a megawatt-hour's price.
This is where the crypto reader's interest actually sits. Bitcoin miners spent the last cycle acquiring two things that AI operators now desperately need: land with substation access and long-dated power contracts. An ASIC that earns $0.06 per kilowatt-hour of work cannot outbid a GPU rack that monetizes the same kilowatt-hour at multiples of that. So miners convert. I watched this at close range โ in early 2024 I helped a small Taipei-linked mining cooperative model the switching cost from ASIC halls to GPU racks, and the arithmetic was brutal in one direction only. The power contract was the asset. The silicon was the depreciating liability. That is why a crypto outlet runs AI infrastructure stories: its audience owns the scarce input.
Political backing has a specific commercial meaning, and it is not subsidy. It is permitting velocity. A data center that clears interconnection and zoning six months earlier starts billing six months earlier, and on a project with a multi-decade payback, six months of earlier revenue is worth more than most tax credits. Code is law, but bugs are reality. The interconnection agreement and the tariff schedule are the law. The bug is the queue โ a physical constraint the contract cannot wish away. Endorsement shortens the queue. It does not lengthen the wire.
Now the contrarian read, and it is the one I would bet on. The important signal is not that Trump defended data centers. It is that the defense was necessary. AI infrastructure has been sliding from bipartisan national asset to partisan cost dispute. The driver is not AI safety. It is distribution. Data centers deliver concentrated benefits โ construction jobs, temporary tax base, shareholder returns โ and disperse costs: higher residential rates, water for cooling, land, noise, and tax exemptions that shift burden to residents. When the benefits are visible and the costs are diffuse, opposition organizes slowly. When the bill arrives on a monthly utility statement, it organizes fast.
I have seen this movie. The mining industry ran the same play โ promise jobs, negotiate tax abatements, absorb cheap power โ and then met the same backlash when residents connected the dots between local load growth and their bills. The pattern is not ideological. It is arithmetic wearing a political costume. Privacy is a feature, not a bug applies here in an unexpected register: the opacity of a data center's actual power draw and tax treatment is a feature of how these deals are sold, and the moment that opacity is pierced by a public utility filing, the political math flips.
There is also a manufactured-narrative layer worth naming. "AI leadership" functions much like "liquidity fragmentation" does in DeFi โ a framing that converts a capex race into a national-security imperative, which conveniently insulates the spend from ordinary cost scrutiny. I have made this argument about cross-chain verification before: when a mechanism depends on an oracle and a relayer, calling it decentralized is a marketing decision, not a cryptographic one. The same discipline applies here. When a data center deal depends on a tax exemption and a capacity queue, calling it "national competitiveness" is a political decision, not an economic one.
The counter-intuitive claim: the bipartisan criticism is not the threat to AI infrastructure. It is the release valve. The genuine danger is the opposite scenario โ that AI data centers remain a bipartisan consensus long enough to lock in cost externalization before residents price it. A consensus that survives scrutiny is one that has already paid for its social license. A consensus that requires a rally to defend it has not.
The second blind spot is the bottleneck. The policy conversation fixates on chips โ export controls, fabrication capacity, advanced packaging. But a GPU you cannot power is a space heater. The binding constraint for US AI expansion is shifting from silicon to interconnection and permitting. That reframes the entire competition. If the US constraint is social license and the grid, then export controls on chips lose marginal effectiveness, and jurisdictions with faster permitting โ the Gulf, parts of Southeast Asia โ inherit an advantage they did not earn through technology. The competitive edge moves from what you can fabricate to what you can energize.
Watch three things. PJM capacity auction clearing prices and residential tariff filings in data-center counties. State legislatures revisiting data center tax exemptions โ Ohio first, because it is the test case. And hyperscaler capex guidance, where the words "power" and "permitting" will start appearing next to dollar figures. If social license breaks, it will not announce itself. It will show up as a delayed project and a revised tariff. Math doesn't negotiate. Neither, eventually, does a voter.