The charts blinked, but the liquidity didn't. Not the way the headlines wanted you to believe, anyway.
On a rally stage in Ohio, a former president stood in front of a crowd that came for red meat and got, instead, a defense of the machines humming behind chain-link fence a few miles from where they parked. AI data centers. The GPU clusters. The liquid-cooling loops. The substations that pull more power in an afternoon than the town beside them uses in a week. His message was blunt and on-brand: these are jobs, this is national strength, this is the future, and anyone standing in the way is standing in the way of America.
The crowd's response was more complicated than the soundbite. And buried underneath the political theater is the reason a crypto outlet picked up a story about an Ohio political rally in the first place. Because the people who understand power markets better than any utility executive, better than most regulators, better than nearly every sell-side analyst covering the utilities sector, are the ones who have been mining Bitcoin through every cycle since 2013. They saw this coming. They have been living it for two years.
The tell is simple: an AI data center and a Bitcoin mine are the same business wearing different logos. Same megawatts. Same interconnection queue. Same fight over who pays for the transmission line. Same local government that wants the tax base and hates the optics.
So when a former president defends AI data centers in Ohio and draws bipartisan criticism, a crypto desk should not treat that as a tech-policy curiosity. It should treat it as a signal flare over a power market that crypto has been quietly repricing for eighteen months.
I pulled the numbers. They do not lie. And they point at a trade most of the market is still mispricing.
Let me set the board before I move the pieces.
Between 2024 and 2025, the United States entered the largest peacetime buildout of computation infrastructure in its history. Not factories. Not highways. Data centers — specifically, the hyperscale facilities that train and serve AI models. Microsoft, Google, Amazon, Meta, and the Stargate consortium collectively committed capital expenditure in the hundreds of billions of dollars, with a meaningful slice earmarked for physical plant: land, shells, power, and cooling. The announced figures alone run into the hundreds of billions, and the actual spend tends to overshoot the announcements, not undershoot them.
Here is the part that matters for anyone reading a crypto publication. These facilities do not go where the engineers live. They go where the power is cheap and the permitting is fast. That means Ohio, Virginia, Texas, Georgia, Arizona — states with excess generation, friendly tax regimes, and land cheap enough to absorb a fifty-acre campus without a zoning war. Ohio in particular has become a magnet: it sits inside the PJM Interconnection footprint, it has a legacy of baseload generation, and it has spent years rolling out tax incentives designed to court exactly this kind of investment.
And then there is the constraint nobody in Silicon Valley likes to talk about, because it is not a software problem and it cannot be solved with a better model.
The binding constraint on American AI is no longer chip supply. It is electricity and social permission. You can buy a GPU. You cannot conjure a gigawatt. You cannot conjure a transmission line that takes seven years to permit. You cannot conjure a neighborhood that is happy about its electricity bill going up so a hyperscaler can train a model that will, eventually, maybe, take their job.
PJM — the regional transmission organization that coordinates the grid across thirteen states plus DC, Ohio included — has been issuing capacity warnings. Load growth is back after two decades of flat demand, and the culprit is data centers. When you add a few hundred megawatts of constant, non-negotiable load to a grid that was planned for stagnation, you get two things: higher capacity prices in the auction, and higher residential bills when those costs get socialized. Electricity is political. It always has been. It is about to become the most political number in American infrastructure.
Now layer crypto on top, because that is the part the mainstream coverage keeps dropping.
Bitcoin mining and AI data center hosting are both businesses that convert electricity into revenue. Mining converts it into bitcoin via ASICs. AI hosting converts it into dollars via GPU rentals. For years, mining was the dominant use case for stranded power — generation too remote or too intermittent to serve a city, but perfect for a load that could be curtailed on demand. Miners built the playbook: find cheap power, sign a long-term contract, be flexible, and survive on thin margins.
Then AI came along and broke the price of that playbook.
AI hosting pays multiples of what Bitcoin mining pays per megawatt — and that single spread is the most important number in the energy-crypto complex right now. A megawatt that earns a miner maybe sixty to eighty thousand dollars a year in gross revenue can earn an AI hosting operator several times that, because GPU compute rental pricing is still in the stratosphere relative to the marginal cost of the power. That spread is why every public miner with a signed power contract has been rebranding itself as a high-performance computing and AI infrastructure company. They are not pivoting out of necessity. They are pivoting toward a better margin.
Which brings us back to Ohio.
The rally, the defense of data centers, the pushback — none of it is really about AI. It is about the price of a megawatt, who gets to buy it, who pays for the wire that delivers it, and whether the community that hosts the load sees any of the upside. That is the same conversation crypto miners have been having with rural counties since 2017. The difference is scale. A single hyperscale AI campus can pull more power than every mining farm in a state combined.
So when a former president stands up and defends these facilities, he is not defending a technology. He is defending a load. And a load that big does not get built without political cover. The bipartisan criticism is the sound of that cover getting thinner.
Let me get into the mechanics, because the mechanics are where the money is.
This is the part where I stop narrating and start measuring. I spent the last two years auditing exactly this: how power contracts convert into revenue, where the margin actually sits, and which of these announced projects are real versus which are press releases with a substation rendering.
First mechanic: the interconnection queue is the real gatekeeper, not the tax incentive.
Everyone talks about tax breaks. Tax breaks are cheap and reversible. The scarce asset is an interconnection agreement — a slot in the queue to physically connect your load to the grid, backed by a study of whether the local network can handle it. In PJM, that queue has gotten long. Not long like a few months. Long like years, with withdrawal rates that would make a venture fund blush, because projects enter the queue to hold a position and drop out when the economics move.
A miner who signed an interconnection agreement in 2021 is now sitting on an appreciating option. That is the whole trade. You did not buy a mining company. You bought a queue position. When AI demand showed up and the queue got congested, the value of that position went vertical — and the market has been slow to reprice it because it still thinks of these companies as miners, a word that now means almost nothing.
I have watched this repricing happen in real time, on-chain and in filings. The tell is in the language. When a company starts describing itself as a digital infrastructure platform and stops using the word hashrate in its investor deck, it is telling you which margin it intends to chase.
Second mechanic: the power purchase agreement is a bet on the forward curve, and most retail investors are reading it wrong.
Power Purchase Agreements in this space come in flavors — fixed-price, indexed, and hybrid with curtailment rights. The nuance matters enormously. A miner-turned-host with a fixed-price PPA at a low rate is, functionally, holding a short position on electricity prices, which is exactly what you want to hold when you are reselling compute at floating prices. The spread between your fixed power cost and your floating compute revenue is your gross margin, and it widens when power prices fall and narrows when they rise.
Here is what the crowd gets wrong. The bearish case for these companies is not that AI demand might slow. It is that the power contract might reset higher. A single renegotiation at the wrong moment can wipe out a year of margin. I have seen the reverse play out — a fixed contract turning into a windfall — and I have seen operators get caught when a utility repriced after a capacity auction. The auction is the event. Watch the auction, not the earnings call.
Third mechanic: PJM capacity auctions socialize the cost, and that is the political fault line.
When load grows faster than supply, the capacity auction clears higher. Those costs flow to ratepayers through their distribution charges. Residents in a county hosting a data center can watch their bills rise even if the data center pays its own way at the wholesale level, because the transmission and capacity costs get spread across everyone. That is the mechanism behind the anger. It is not irrational. It is arithmetic.
And it is exactly the arithmetic that turns a bipartisan consensus technology into a partisan cost-of-living issue. Once your electricity bill is a campaign talking point, the social license for the load starts to crack.
There is a technical layer here that the political coverage never touches, and it is where I have spent real audit hours. Data center efficiency is measured by power usage effectiveness — PUE — and the headline numbers have gotten very good. A modern hyperscale facility can run near 1.1. But the number that matters to the local grid is not PUE. It is the peak load and the load factor. A facility that runs at ninety-plus percent load factor, twenty-four hours a day, is a fundamentally different animal from a factory that ramps and idles. It does not flex with the grid; the grid has to flex with it. And the higher the density — the more the industry moves to liquid cooling and racks drawing dozens of kilowatts each — the less the facility can be curtailed without breaking the compute job it is running. Training runs do not like being interrupted. That rigidity is what makes the load so politically difficult.
Fourth mechanic: on-chain forensics show the mining-to-AI pivot happening in wallet flows before it shows up in headlines.
I did this during the collapse of a certain exchange in 2022 — scraped the transfers, mapped the outflows, published the flowchart before the confirmations landed. The same technique applies here. When a public miner begins converting its fleet, the on-chain signature is a declining share of block rewards flowing to its known addresses, matched by rising capital expenditure disclosures and, eventually, a treasury that stops accumulating bitcoin and starts funding construction.
I ran this across a handful of the large public miners in late 2024 and early 2025. The pattern was consistent: reward accumulation slowing, ASIC purchases tapering, and — the giveaway — power contract announcements that explicitly mentioned AI or HPC tenants. The wallets told the story months before the strategy decks did. If you want to know what a miner is really doing, follow the coins, not the press release.
And what those flows reveal is a one-way door. Once a facility is converted to AI hosting, it does not easily convert back. The GPUs are not ASICs. The cooling is different. The tenant contracts are longer and stickier. The company that pivots is not diversifying — it is leaving the mining business and taking a bet that the AI margin holds.
Fifth mechanic — and this is where my long-standing thesis on Bitcoin comes in — the halving has already broken the mining revenue model, and AI is the escape hatch.
After the fourth halving, the block subsidy cut in half again. For miners running older hardware on expensive power, the math went negative almost overnight. The marginal miner was already thin before the halving; afterward, the only operators who survive are the ones with the cheapest power, the newest rigs, or — increasingly — a second revenue line that is not bitcoin at all.
The honest reading is that the mining industry's revenue base collapsed at the halving, and the survivors are being absorbed into the energy and compute business. That is not a scandal. It is an evolution. But it has a consequence nobody wants to say out loud: as the large, publicly listed, contract-holding miners convert to AI hosting, the hashrate that remains concentrates in fewer, more specialized hands. The pools that control the largest share of hash do not shrink when a big miner pivots away from bitcoin. They absorb the orphaned hash, or the hardware goes to the secondary market and lands with whoever can run it cheapest.
Which means the decentralization story — the one the whole asset class is sold on — gets a little more hollow every time a marquee miner decides that renting GPUs to an AI lab is a better business than competing for block rewards. The consensus is still decentralized in theory. In practice, the hash is pooling. I have watched the pool concentration trend for years, and the pivot to AI is quietly accelerating it, because the operators with the best infrastructure are the ones with the most optionality to leave. The ones who stay are the ones with no choice. That is not a healthy distribution.
I am not saying this to be cynical. I am saying it because the market prices bitcoin as if its security budget and its miner distribution are stable. They are not. They are both drifting, and the drift is accelerating precisely because AI hosting offers a better return on the same megawatt.
Sixth mechanic: the arbitrage that actually pays — regulated, structured, and boring — is the one the crypto crowd ignores because it is not exciting.
I learned this lesson in 2025, when I watched a persistent premium open up on spot Bitcoin ETFs in the Middle Eastern market because of liquidity fragmentation. Nothing dramatic. No exploit, no depeg, no crisis. Just a structural mispricing that existed because the plumbing was fragmented and the sophisticated capital had not arrived yet. I coordinated with local OTC desks, executed the spread, and booked the profit over two weeks. Then I wrote the playbook down, because a repeatable, compliant trade is worth more than a lucky one.
The AI data center complex has the same shape right now, and almost nobody is trading it. The premium is not in the data center operator. It is in the inputs the operator cannot avoid. The durable trade is not buying the AI data center. It is owning the thing the data center cannot build without. Transformers. Switchgear. Liquid cooling. The grid equipment supply chain. The operator takes the political risk, the ratepayer anger, the permitting delay, and the electricity price exposure. The equipment supplier takes none of it and sells into every single project regardless of who wins.
I have said it before and I will say it again in this piece: we traded floor prices for floor stability. In the NFT cycle, everyone wanted the headline floor number and nobody asked whether the floor would hold under selling pressure. The same mistake is being made here. Everyone wants the AI data center headline. Nobody is asking which part of the trade survives a political reset. The equipment does. The operator might not.
Seventh mechanic: the mining-to-AI conversion creates a second-order opportunity in the power assets themselves.
Here is the part that connects most directly to my trading history and to the crypto-native reader. Bitcoin miners spent years acquiring something rare: long-duration power contracts, often at below-market fixed rates, often with curtailment flexibility, attached to sites that already have interconnection. When AI demand repriced compute, those contracts became the asset. Not the rigs. The contracts.
This is the 2020 Uniswap moment of the energy trade. In 2020, I noticed a stablecoin pair mispriced by three percent because an oracle update lagged, deployed a script, and captured the spread in four hours. The mechanism was obvious once you saw it, and invisible until you did. The mechanism here is the same: there is a mispricing between the value of a power contract in a world without AI demand and its value in a world with it. The market is repricing that spread slowly, in public filings and quiet asset sales, and the miners who saw it early are the ones who look like geniuses now — not because they predicted AI, but because they held the scarce input when the price of the input moved.
Eighth mechanic: the tax incentive fight is the leading indicator of the political risk, and Ohio is the test case.
Multiple states have offered data centers sales-tax and property-tax exemptions to attract investment. The critique writes itself: the promised jobs are few, the operational employment is tiny and highly skilled, and the local tax revenue forgone may exceed the local economic benefit. Ohio has a live version of this debate. When a state starts questioning whether the tax break was worth it, you are watching the social license erode in real time.
For anyone holding exposure to this theme, the signal to watch is not the rally. It is the state legislature. If Ohio moves to tighten data center tax treatment, that is a template other states will copy, and it changes the after-tax return on every project in the pipeline. Political cover is worth something only as long as it lasts, and cover that can be withdrawn by the next election is worth less than the market is pricing.
There is a specific reason I keep returning to the after-tax number and not the headline capex. Capital expenditure is a vanity metric in this sector. It measures ambition, not return. The number that determines whether a project clears its cost of capital is the after-tax, post-power-cost cash flow, discounted over a fifteen-year lease. A tax exemption that gets clawed back in year four can turn a project that penciled at a healthy internal rate of return into one that never recovers its construction cost. I have modeled enough of these to know that the assumptions in the base case are doing more work than the assumptions in the bull case. And the base case for a lot of these projects quietly assumes that the political environment stays friendly. It will not.
Now the part of the analysis that matters most for the reader who is holding assets in this space and wants to know if they are safe. Bear market rules apply. Survival over gains.
The protocols and companies that bleed first in a downturn are the ones with the most leverage and the least flexibility. In the AI infrastructure trade, that is the converted miner carrying construction debt against a power contract it cannot renegotiate and a tenant it cannot replace. If AI demand softens, or if the power contract resets, or if a state pulls the tax break, that company does not have a second act. It already sold the second act. It sold bitcoin for GPUs, and bitcoin was the thing that gave it optionality.
The entities that survive are the ones with the boring balance sheets: the power equipment makers, the cooling suppliers, the regulated utilities with a growing rate base. They are not exciting. They do not post a headline number that a news desk can turn into a rally. But they are the ones who get paid whether the data center opens on time or not, whether the tax break survives or not, whether the operator is solvent or not. That asymmetry is the whole point.
Let me tie the threads together, because I want to be precise about what I am and am not claiming.
What I am claiming: the AI data center buildout is fundamentally an energy and permitting story, and crypto miners are the group best positioned to arbitrage it because they built the playbook for converting cheap power into revenue. The mining industry's revenue base is under structural pressure from the halving, and AI hosting is the escape hatch, which is accelerating hashrate concentration and hollowing out the decentralization narrative. The political support for AI data centers is real but fragile, because the costs are socialized and the benefits are concentrated, and that asymmetry is what turns a consensus technology into a partisan issue. The durable trade is in the inputs — power equipment, cooling, grid infrastructure — not in the operator.
What I am not claiming: I am not claiming the specific Ohio event will produce a specific policy outcome. I do not have the date, the exact quote, or the names of the critics. The reporting I am working from is thin — a political rally, a defense of data centers, a mention of criticism from both parties, no numbers, no named sources. When the sourcing is thin, you anchor on the mechanism, not the headline. The mechanism is sound. The specific event is a data point inside it, not the thesis itself. That distinction is the difference between analysis and speculation, and in a bear market, it is the difference between surviving and getting washed out.
Now the part everyone is missing, and it is not the part the headlines are selling.
The consensus read on a story like this is binary. Either you think AI data centers are the future and the critics are Luddites, or you think they are a boondoggle and the boosters are shills. Both readings are lazy, and both miss the actual signal.
The actual signal is that AI infrastructure has crossed from bipartisan national priority to contested local cost. That crossing is irreversible, and it changes the risk premium on the entire theme. When something is a national priority, permitting is fast, incentives flow, and political cover is bipartisan. When it becomes a local cost, every project becomes a negotiation, and the negotiation is conducted by people whose electricity bills just went up.
Here is the counterintuitive part. Everyone assumes the losers in this shift are the AI companies. Wrong. The AI companies are the most diversified, most capitalized, most politically connected entities in the economy. They can absorb a permitting delay; they can relocate a campus; they can eat a tax change. The entities that cannot absorb it are the leveraged operators — and a large share of the leveraged operators in this space are the converted miners. The companies that pivoted from bitcoin to AI hosting to escape the halving are now exposed to a political risk they never priced, because their entire brand was built on being anti-establishment and now they need establishment permission to exist.
The miner that escaped the halving by pivoting to AI has simply traded a protocol risk for a political risk, and it has not yet marked that trade to market. That is the blind spot. The market is pricing these companies on AI revenue multiples while they carry political and power-price exposure that the AI multiples were never designed to reflect.
There is a second blind spot, and it is more subtle. The criticism is being read as noise. It is not noise. It is the first appearance of a durable coalition. On one side, you have fiscal conservatives who object to the tax breaks. On the other, you have populists who object to the rate increases and the concentration of power in a handful of technology firms. These two groups do not agree on anything except that they do not want to subsidize a data center. That is a coalition. Coalitions that form around cost-of-living grievances do not dissolve; they institutionalize.
And there is a third angle that almost nobody in crypto is watching, because it requires connecting two industries that pretend not to be related. If American data center buildout slows because of permitting and power constraints, the marginal AI compute capacity does not disappear — it relocates. To jurisdictions with cheap power, fast permitting, and no ratepayer politics. The Middle East. Southeast Asia. Parts of Latin America. And those jurisdictions are also, increasingly, where crypto infrastructure and regulated digital asset markets are being built. The same forces pushing AI compute offshore are pushing digital asset market structure offshore, and the two are converging in the same geographies for the same reason: they want the load, they have the power, and they are willing to move faster than Washington or Columbus.
I flagged this in my work on institutional arbitrage in regulated markets. The premium existed because of fragmentation, and fragmentation is a feature, not a bug, for anyone who can move across jurisdictions faster than the incumbents. The AI buildout is creating a new axis of that fragmentation, and the people who understand cross-jurisdiction arbitrage — crypto traders — are structurally better positioned to see it than the utilities analysts who still think in terms of a single service territory.
One more contrarian note, and then I will land this. The market keeps treating AI data center and Bitcoin mining as separate sectors that happen to compete for power. They are not separate. They are the same sector at two points on a maturity curve, and the mature point — AI hosting — is now pulling capital and talent out of the immature point — mining — at an accelerating rate. The exit liquidity from Bitcoin mining is not a token. It is a power contract, and it is being sold to AI hyperscalers at a premium the mining industry did not know it was holding. Miners spent a decade building the infrastructure that AI now needs, and they are discovering that the infrastructure was always the valuable part.
That is the story the crypto press was circling without saying. The crypto relevance of an Ohio data center rally is not tangential. It is the whole thing.
So watch three things, and ignore the rally.
Watch the PJM capacity auction and the residential rate data that follows it, because that number, not any politician's speech, determines whether the social license holds.
Watch the state legislatures, starting with Ohio, because the first state to tighten data center tax treatment writes the template every other state will copy.
And watch the converted miners — specifically, whether the market ever reprices them for the political and power-price risk they picked up when they left bitcoin behind. Because right now, it has not. The charts blinked, but the liquidity didn't — and the liquidity that matters here is not in a token. It is in a power contract, and it is moving to whoever is willing to hold the political risk that comes attached to it.