The queue is the tell. Not the chart. Not the funding rate. The queue.
Over the past ninety days I have been pulling public interconnection filings from the three largest grid operators in the United States โ ERCOT, PJM, and MISO โ and cross-referencing them against the power purchase agreements that listed miners disclose in their 10-Ks and 10-Qs. The pattern is not subtle. New large-load interconnection requests are being filed at a rate the transmission planning process was never engineered to absorb, and the political system has started to notice. That notice is arriving in the form of moratorium proposals โ not federal, not coordinated, but state-by-state, county-by-county, and in some cases municipality-by-municipality.
Here is the part the headlines miss. A moratorium is not a price signal. It is a supply constraint. And in a market where the supply of permitted megawatts is the binding constraint on hashrate growth, a supply constraint on permits is functionally a supply constraint on hashrate itself. The vote has not happened. The repricing already has.
Let me back up, because the framing matters more than the noise.
Context: What Is Actually Happening, and Why Now
The story circulating in crypto media is simple and, on its surface, alarming: global opposition to data centers is rising, and US states face potential moratoriums. Five information points. No named state. No bill number. No timeline. The word "may" doing most of the structural work.
If you trade on that alone, you are trading on a vibe. I do not trade vibes. I trade constraints, and the constraint here is legible once you stop reading the headline and start reading the grid.
Data centers โ and this is the crux โ are not one industry. They are two industries wearing the same coat. The first is AI compute: hyperscale GPU training and inference clusters, the kind that Nvidia's order book and Microsoft's capital expenditure guidance are built on. The second is proof-of-work mining: ASIC clusters converting electricity into hashrate into bitcoin. These two industries are physically near-identical at the facility level. Both need large contiguous land parcels. Both need high-voltage interconnection. Both need liquid cooling at scale. Both need water for evaporative cooling, or an expensive air-cooled workaround. Both produce heat, noise, and a very visible footprint in the communities that host them.
The convergence is not a metaphor. It is a physical fact, and it is the reason this story belongs in a crypto feed at all. When a county commission votes to pause new "high-energy-use facilities," the language is deliberately technology-neutral. It catches the AI campus and the mining farm in the same net. The mining farm just has less political cover to escape it.
Why now? Three curves crossed.
First, the AI capital expenditure curve went vertical. The buildout of training clusters created a step-function increase in large-load interconnection demand that grid planners, whose models were calibrated on two to three percent annual load growth, simply did not forecast. When your load forecast is off by an order of magnitude, your transmission plan is worthless, and the queue backs up.
Second, the residential electricity price curve turned upward in exactly the jurisdictions where the buildout was concentrating. Ratepayers noticed. Ratepayers vote. Utility commissions, which are appointed or elected and acutely sensitive to the price of a kilowatt-hour on a household bill, began to treat large-load interconnection as a political liability rather than an economic development win.
Third, the mining curve, post-halving, forced operators to compete on cost in a way they never had to before. The April 2024 halving cut the block subsidy, compressed margins, and turned every basis point of power cost into an existential variable. That pushed miners to chase the cheapest stranded power on earth โ which, inconveniently, tends to be located in the same places the AI buildout wants, for the same reasons.
Three curves, one resource. Power, land, and social license. This is the collision.
Core: Reading the Constraint, Not the Headline
The physics of the collision
Start with the unit economics, because everything downstream is a consequence of them.
A modern air-cooled ASIC โ take the Antminer S21 class as the reference point โ draws somewhere in the neighborhood of 3.5 kilowatts and produces roughly 200 terahashes per second. Multiply that across a 100-megawatt facility and you get a machine count in the tens of thousands and a hashrate contribution measured in the low exahashes. That is the physical substrate. It is not abstract. It is steel, silicon, and a substation.
Now put an AI training cluster next to it. A single rack of eight top-tier GPUs can pull north of 10 kilowatts on its own, and a training hall with thousands of GPUs draws tens of megawatts โ before you count the cooling overhead, which, depending on whether you are running air or direct-to-chip liquid, adds another twenty to forty percent. The power usage effectiveness number, the ratio of total facility power to compute power, is the number operators live and die by, and at the AI end it is a genuine engineering problem rather than a rounding error.
Here is the point that gets lost. Both facilities are, from the grid's perspective, the same load. They are large, they are firm, they are constant, and they are connected at transmission voltage. The grid does not care whether the electrons are hashing a block or multiplying a matrix. It cares about the megawatt, the ramp rate, and the reactive power characteristics. On those dimensions, the mining farm and the AI campus are twins.
That is why the regulatory net, when it drops, drops on both.
And this is where the first uncomfortable insight lives: the crypto industry spent a decade arguing that mining is uniquely flexible, uniquely interruptible, uniquely grid-friendly because it can curtail on command. That argument was true and it was under-monetized, and now the AI industry is making the same argument with better lawyers and a national-security framing attached. The miner's curtailment story is competing against a hyperscaler's curtailment story, and the hyperscaler is telling it in front of a congressional committee while wearing a flag pin.
The regulatory transmission chain
The source material gestures at a chain without spelling it out, so let me spell it out, because the chain is the trade.
It runs like this. Social opposition โ noise complaints, water withdrawal permits challenged, ratepayer anger over utility bills โ produces regulatory scrutiny at the state public utility commission level. That scrutiny produces legislative attention, because legislators read the same angry constituent mail the commissioners do. Legislative attention produces bills, and bills produce moratoriums, either as a blunt instrument (pause all new permits of a given class for a defined period) or as a scalpel (tighter water rules, higher minimum efficiency standards, mandatory demand-response enrollment, or a new tariff class for large loads).
This is a bottom-up, socially driven pathway, not a top-down federal one. That distinction is the whole ballgame, and it is why you should not be looking at Washington for the signal. You should be looking at statehouses and county courthouses.
I have seen this movie before, and I watched it from the inside. When the EU's MiCA framework came into force and I went hunting for the arbitrage, the thing that surprised me was not the headline rules โ those were published and legible. It was the reserve requirement loopholes in the balance sheets of three major stablecoin issuers, the shadow-banking channels that let them satisfy the letter of the capital rules while sidestepping the spirit. Regulators do not write the rules that bind; they write the rules, and then the regulated entities find the seam. Data center moratoriums will work the same way. The blunt pause is the headline. The seam โ the efficiency carve-out, the demand-response exemption, the legacy-permit grandfather clause โ is where the actual capital will flow.
Watch for the seam. It always exists.
The interconnection queue as the leading indicator
If you want one number to track, it is not the price of bitcoin and it is not the hashrate chart. It is the interconnection queue depth and the projected completion dates attached to it.
Large-load interconnection is a queue process. You file, you study, you wait, and in the mature markets the wait is now measured in years โ often four to seven โ because the transmission upgrades required to serve you have to be planned, permitted, financed, and built before you can energize. This is the physical pipeline of future compute. It is the leading indicator that precedes every hashrate print and every AI cluster announcement by half a decade.
So when a state proposes a moratorium, what it is really proposing is to freeze that queue. Existing projects with executed interconnection agreements and signed large-load service contracts are, in most draft language I have reviewed, grandfathered. New entrants are not. The effect is to convert the queue from a competitive race into a moat.
This is the single most under-reported consequence of the moratorium wave, and I will come back to it, because it inverts the naive bearish read.
The political economy: why miners have no cover
Let me be blunt about the asymmetry, because it is the core of my contrarian thesis and it deserves to be stated plainly rather than buried.
The AI industry has three forms of political cover that the mining industry does not.
It has a national-security narrative. Compute is treated, increasingly, as a strategic asset in a technology competition with China. A hyperscaler can walk into a statehouse and say, credibly, that its facility is part of the national AI infrastructure. A mining farm cannot make that claim, because it is not true, and everyone in the room knows it.
It has a jobs narrative that survives scrutiny. An AI campus employs a meaningful number of high-skill, high-wage workers in operations, networking, and facilities engineering. A mining farm employs, at the same megawatt scale, an order of magnitude fewer people, most of them in roles that do not require a local talent pipeline. The economic development multiplier is simply different, and the community impact statement reflects it.
It has a diversified lobbying footprint. The hyperscalers and their supply chain โ the chipmakers, the equipment vendors, the cloud providers โ spend across every level of government, and they spend as a bloc. Mining lobbying is comparatively thin, geographically concentrated, and burdened by a decade of bad press that it never fully shed.
So when the net drops, the miner is the one without the umbrella. The moratorium language is neutral. The enforcement pressure is not.
And here is the second-order effect nobody is pricing: because the AI industry needs the same permits the mining industry needs, and because the AI industry has the political capital to negotiate carve-outs, the likely outcome is not that all large loads get frozen. It is that the right kind of large load gets expedited and the rest gets stalled. Mining is the rest.
The pivot economics: from hash to host
This is where the real capital is moving, and it is why I have been repositioning my own watchlist for six months.
The most important corporate development in the mining sector over the past two years has not been a hashrate record. It has been the migration of listed miners into high-performance computing and AI hosting. The thesis is straightforward: a miner already owns the land, the interconnection, the substation, and the shell. Converting a portion of that capacity from ASIC hosting to GPU hosting monetizes the exact assets that are becoming scarce โ permitted megawatts โ at a much higher revenue per megawatt than pure bitcoin mining can support at current hashprice.
Run the numbers and the gap is not marginal, it is structural. The revenue per megawatt of an AI hosting contract, priced on a multi-year take-or-pay basis, can run several multiples of what the same megawatt earns hashing bitcoin at a compressed subsidy. That is not a comment on bitcoin's long-term value. It is a comment on the fact that a hyperscaler, desperate for energized capacity that will not take five years to deliver, will pay a premium for someone else's place in the queue.
This is the seam. The moratorium accelerates the pivot, because the moratorium makes the queue position โ the permit, the energized capacity, the executed interconnection agreement โ the scarce asset, and the pivot is precisely a strategy for monetizing that scarcity.
I want to be careful about the arithmetic here, because the bull case oversells it. The conversion is not free. GPU hosting requires different cooling, different power distribution, different networking, and a customer relationship and operational competency that a mining operator may not possess. The capital expenditure to retrofit a mining hall into a Tier III-equivalent AI facility is substantial, and the execution risk is real. Some operators will botch it. Some will sign hosting contracts they cannot physically deliver against, and that will be a story in eighteen months. But the directional logic โ permitted megawatts are the asset, and the asset is being repriced upward by scarcity โ is sound, and it is the correct frame for the entire moratorium discussion.
Case studies: the precedents that already exist
The source material mentions moratoriums as a possibility. They are not hypothetical. The precedents exist, and the shape of the precedent tells you the shape of what comes next.
New York is the canonical example. The state enacted a moratorium on new fossil-fuel proof-of-work mining permits, the first of its kind, and the mechanism is instructive: it did not ban mining. It froze the permitting pathway for a specific class of facility for a defined period while the state studied the environmental impact. That is the template. A targeted, time-boxed freeze on the permit pipeline, dressed in environmental language, justified by a study that will take longer than the moratorium itself.
You can see the same DNA in the noise ordinances that have spread through mining-heavy counties. The mechanism there is different โ a decibel cap rather than a permit freeze โ but the effect is identical: it raises the cost of new entry and protects incumbents who were already operating under the old rules.
Texas is the counter-example and the pressure valve. ERCOT's market structure, with its energy-only design and its aggressive demand-response programs, made the state the preferred destination for mining load through the last cycle. But Texas is not immune. The same interconnection queue congestion that afflicts PJM is building in ERCOT, and the same ratepayer politics that drove the New York action exist in Austin. The difference is that Texas has, so far, chosen to monetize flexibility โ paying large loads to curtail โ rather than to restrict entry. That choice is the state's competitive advantage, and it is also the thing most likely to be politically tested if residential bills keep climbing.
Georgia, Virginia, Ohio โ the pattern repeats with local variation. What matters is not the specific state but the structural fact: the policy is moving from a posture of welcoming large load to a posture of scrutinizing it, and the scrutiny is not evenly distributed across facility types.
The moat effect: why a moratorium can be bullish
Now the inversion. Stay with me, because this is the part that separates the reflexive bear from the operator.
The consensus read of a moratorium is bearish for mining: less new capacity, slower hashrate growth, worse economics for the sector. That read is correct about the sector and wrong about the incumbents.
Think about what a moratorium actually does to the competitive structure. It caps the supply of new permitted megawatts. For an operator with existing permits and energized capacity, that cap is a gift. It raises the replacement cost of the asset they already own, it removes the threat of a flood of new entrants competing for the same power contracts, and it strengthens their pricing power in any hosting negotiation, because the counterparty's alternative is to build from scratch into a queue that may be frozen.
This is the same dynamic that plays out in every capital-intensive industry when the permit becomes the bottleneck: the value migrates from the operator to the permit holder. The permit is the asset. Everything else โ the ASICs, the racks, the software โ is commoditized and replaceable. The interconnection agreement is not.
So the correct framing is not "moratoriums are bad for miners." It is "moratoriums are bad for marginal miners and good for entrenched ones." The dispersion of outcomes widens. The pure-play, single-site, single-jurisdiction operator is exposed. The diversified, multi-site operator with a portfolio of executed interconnections is insulated, and may even benefit. This is a stock-picker's regime, not a sector trade.
Demand response: the mitigation that pays
The mechanism that mediates all of this is demand response, and it deserves its own section because it is where the mining industry's oldest argument finally finds its market.
A large load that can curtail on command is not just a load. It is a grid asset. In an energy-only market with high renewable penetration, the value of a flexible, fast-responding, multi-hundred-megawatt load that can drop to zero in seconds is substantial. It substitutes for peaker plants, it absorbs renewable oversupply that would otherwise be curtailed, and it stabilizes frequency.
Miners have been selling this story for years. What is new is that it has become a licensing condition. Draft large-load tariff language increasingly conditions new interconnection on mandatory demand-response enrollment. You want the megawatts? You commit to shedding them on the system operator's schedule, not yours.
This changes the economics of a mining facility in a way that most models do not capture. A facility that is curtailed for a meaningful number of hours per year has a lower effective utilization, which lowers its effective hashrate output, but it also earns curtailment payments and it accesses power contracts that a non-flexible load cannot. The net present value calculation is genuinely ambiguous, and it depends on the shape of the local market and the terms of the curtailment compensation.
The operators who understand this are already building it into their site selection. The ones who do not are going to discover it the hard way, when their new facility is denied interconnection because they refused to sign a flexibility agreement.
And here is a genuinely contrarian observation about the energy market that most crypto people miss: the marginal cost curves in electricity are real. Unlike the interest rate models that DeFi lending protocols deploy โ which, and I say this as someone who has watched the mechanics closely for years, are essentially arbitrary administrative choices dressed in algorithmic clothing, with utilization curves that have no demonstrable relationship to any underlying supply-and-demand equilibrium โ the price of a megawatt-hour at four in the afternoon on a hot August day is a discovered price, backed by physical scarcity and verifiable marginal cost. That is why energy is such a powerful anchor for compute economics. The signal is real. The DeFi lending rate is a guess with a governance token attached.
The hashrate redistribution math
Zoom out to the network level and the moratorium wave becomes a geographic reshuffling rather than a reduction.
Hashrate is mobile in a way that most industrial capital is not. The ASICs can be unracked, crated, and shipped. The binding constraint is not the machine, it is the energized, permitted site. So when a jurisdiction tightens, the marginal hashrate does not disappear โ it relocates, over a period of quarters to a couple of years, to the jurisdictions that remain open.
The migration map has a few well-worn destinations. Domestically, Texas and Wyoming remain the friendly jurisdictions, though both are under increasing queue pressure. Internationally, the Middle East โ with sovereign capital explicitly courting compute infrastructure and electricity priced at a fraction of Western rates โ has become a genuine destination rather than a talking point. The Nordics retain an edge on cheap hydro and, critically, on cooling economics, since ambient temperature does a lot of the thermal work for free. Latin America is emerging, with stranded hydro and a growing appetite for dollar-denominated infrastructure investment.
The long-run consequence is a slow erosion of the United States' share of global hashrate, not because of any single moratorium but because of the aggregate effect of many. This matters for network security in a way that is easy to overstate in the short run and easy to understate in the long run. A network whose hashrate is concentrated in a smaller number of jurisdictions, and in jurisdictions with different regulatory regimes, has a different resilience profile than one spread across many. That is a structural consideration, not a price catalyst.
I will say clearly what the source material is too thin to say: none of this is a tradeable event on its own. There is no named state, no bill number, no vote date. It is background noise with a directional tilt. Treating it as a catalyst is a category error. But treating it as irrelevant is a different error, because the direction is real and the second-order effects compound.
The GPU competition nobody is pricing
There is a resource competition inside the compute stack that deserves a paragraph, because it touches a domain I follow closely.
Zero-knowledge proof generation and AI inference are both GPU-bound workloads. They compete for the same silicon, the same data center capacity, and the same energized megawatts. And the economics of the two are wildly divergent. AI inference is a product with customers paying real money today. ZK proving, for most of the deployments I have audited, is a cost center whose economics only work if gas prices are high enough to make on-chain verification worth paying for โ which, at current gas levels, they are not.
So when you add data center capacity constraints and moratorium risk on top of that, you get a compounding squeeze on the proving-cost side. The facilities that could host proving infrastructure are being bid away by AI workloads that can pay more per kilowatt-hour, and the proving operators are left with structurally higher costs and structurally lower revenue. The proving costs are already absurd. Constrained capacity makes them worse.
This is a second-order effect of the data center story that almost nobody is connecting, and it is the kind of thing that only shows up when you have actually tried to run a proving workload and seen the invoice.
The DAO comparison, and why it matters here
The last structural piece is about how the carve-outs get allocated, because that is where the money will actually be decided.
When a state or county negotiates a community benefit agreement โ the package of payments, local hiring commitments, and infrastructure contributions that buys a large load its social license โ it is, functionally, running a grant program. And most grant programs, as I have argued for years and will keep arguing, are captured. The allocation goes to whoever has the relationship, the lobbyist, the local political connection. It is nepotism wearing a process costume.
There is exactly one public-goods funding mechanism I have seen that reliably escapes this, and it is the one that does not rely on a committee making discretionary judgments about who deserves what. It relies on a mechanism that funds what the community actually uses, verifiably, without a gatekeeper deciding in advance. The lesson for data center policy is direct: the carve-outs that get allocated through discretionary negotiation will be captured, and the carve-outs that get allocated through a transparent, usage-based, verifiable mechanism will not. Watch which mechanism the states choose. That choice, more than the moratorium itself, determines who wins.
Contrarian: The Bear Case Is Backwards, and the Bull Case Is Overrated
Everyone is reading this story the same way, and the consensus read is wrong in both directions.
The bearish consensus is that moratoriums hurt miners. Wrong at the operator level. They hurt marginal miners and they help entrenched ones, because the permit is the scarce asset and scarcity of permits is scarcity of competition. The correct trade is not short the sector, it is long the operators with the deepest portfolio of executed interconnection agreements and short the ones whose entire thesis rests on a single greenfield site in a politically exposed jurisdiction. The dispersion is the trade, not the direction.
The bullish consensus, meanwhile, oversells the pivot to AI hosting as if it were free money. It is not. The retrofit capital expenditure is real, the operational competency is not guaranteed, and some of the hosting contracts being announced today will not be delivered against on the timeline promised. When that becomes visible โ and I would look for the first cracks twelve to eighteen months out โ the market will re-rate the pivot story sharply. The operators who actually built the operational muscle will be fine. The ones who sold a narrative and a PowerPoint will not.
The genuinely unreported angle, the one I would put capital behind, is this: the moratorium wave is a permit revaluation event, and the market is repricing the wrong asset. Everyone is watching hashrate and hashprice. The thing being repriced is the interconnection agreement, and that repricing is invisible on any dashboard because it does not trade. It shows up in the premium paid in hosting negotiations, in the valuation multiple assigned to energized capacity, and in the replacement cost of a megawatt that used to be easy to permit and now is not.
The blind spot is that the entire discussion is framed as a policy story. It is not. It is a scarcity story that happens to be enforced through policy. Frame it as scarcity and the positioning becomes obvious.
Takeaway: What I Am Watching Next
Three signals, in order of importance.
The first is named-state legislation. The moment a specific state introduces a specific bill with a specific timeline, the abstraction becomes an event, and the operators exposed to that jurisdiction re-rate. Until then, this is background.
The second is the interconnection queue. If new large-load requests start getting rejected or deprioritized in a major market โ not paused pending study, but denied โ then the permit scarcity thesis is confirmed and the moat widens.
The third, and the one I trust most, is the hosting contract pricing. When the next round of miner-to-AI hosting deals gets signed, look at the revenue per megawatt. If it is climbing, the permit is being repriced in real time, and the whole moratorium conversation is downstream of a market that has already made up its mind.
The chart has not broken yet. But the queue has, and the queue is where the next move is already being priced. Chasing the alpha while the market sleeps means reading the filing nobody is reading, not the headline everybody is.
The endgame here is not a ban. It is a revaluation. And the operators who understood that before the vote will be the ones holding the permits when the music stops.