There is a particular kind of signal that arrives dressed as noise. On an ordinary afternoon in the American South, a cryptocurrency publication โ a venue built to track token prices and protocol upgrades โ quietly published a sixty-word brief about a utility company warning a cloud computing giant that it might cut its power. No token was named. No chain was upgraded. No yield farm was launched. And yet, in that misfiling, I recognized one of the most important stories of the year.
Entergy Mississippi had told Amazon Web Services, in effect, that the electricity promised to its data centers might not be there when the servers were ready to draw it. The phrase used was "potential disconnections." It sounds like a technical footnote. It is closer to a confession. When a utility publicly threatens to disconnect its largest new customer, it is not describing an engineering accident. It is describing a negotiation that has reached the edge of a cliff.
Follow the money, not the noise. And here the money is not in the servers. It is in the copper, the gas, and the transformers that must exist before a single inference can be served โ or a single block can be mined.
I have spent the better part of two decades reading the fine print of systems that promise more than they can deliver. In 2017, I reverse-engineered the smart contracts of a failed payment protocol and watched governance structures turn into liquidity traps within weeks. The lesson was not that the code was bad. The lesson was that technology without an honest accounting of its resource base is a story waiting to collapse. That is precisely what the Entergy brief is about โ except this time the resource is not a token. It is electricity.
The context matters, because without it the brief reads as a local dispute. It is not. The United States is living through the first structural tightening of its electricity system in a generation. For roughly twenty years, American power demand was almost flat โ a plateau so stable that planners built their entire discipline around the assumption of zero growth. Then, beginning around 2023, demand started to climb again, driven by three forces arriving at once: the re-shoring of heavy manufacturing, the electrification of transport and heating, and above all the explosive build-out of data centers to feed the artificial intelligence boom. Forecasters at the Energy Information Administration and the North American Electric Reliability Corporation have repeatedly revised their load projections upward. The plateau is over.
Into this newly hungry system stepped the large load โ a customer so big that a single facility can consume more power than a mid-sized city. The defining large loads of this era are AI data centers and, before them, Bitcoin mining farms. Mississippi, historically a state with surplus power and some of the lowest electricity prices in the nation, had spent years courting exactly this kind of customer. Amazon committed roughly ten billion dollars to the state. The logic was simple and, on paper, irresistible: bring in massive load, expand the rate base, spread fixed costs across more ratepayers, and keep prices low. It is the same logic that once drew miners to places like upstate New York and the plains of Texas.
That logic has a hidden assumption buried inside it โ that the power will actually be there. The Entergy warning is the sound of that assumption cracking.
The first thing to understand is that this is a story about capacity, not about intentions. A data center is an unforgiving customer. It does not tolerate a brownout. It does not reschedule. Its load is flat, dense, and relentless โ seven days a week, twenty-four hours a day, at a power density per square foot that no ordinary commercial building approaches. To serve it, a utility needs what engineers call firm, dispatchable capacity: generation that can be turned on whenever the customer flips the switch. In the hierarchy of American new generation, that hierarchy is now brutally clear. In the short term โ roughly 2025 to 2028 โ natural gas combined-cycle plants will supply the overwhelming majority of new dispatchable capacity, north of eighty percent by most estimates. In the medium term, existing nuclear plants will be extended in life, and small modular reactors will begin to appear. Renewables paired with storage, for all their moral appeal, cannot yet carry the firm-power load that a hyperscale data center demands; they are a supplement, not a spine.
This is where the brief's silence becomes eloquent. Entergy did not specify which generation it lacked. It did not need to. The shortage of firm capacity is now a national condition, not a local one.
Consider the bottlenecks, because they are what transform a routine expansion into a crisis of timing. The delivery lead time for a gas turbine โ the very machine that would most likely power the next wave of data centers โ has stretched from a historical eighteen months to somewhere around 2027 or 2028, as order books at GE Vernova and Siemens Energy have filled beyond their factories' reach. Transformers, the unglamorous steel-and-copper devices that step voltage up and down, now take two to three years to arrive. And the interconnection queue โ the line every new generator must join before it can legally deliver power to the grid โ has swollen to roughly 2,600 gigawatts of proposed projects, with an average wait of about five years. A queue that large is not a backlog. It is a dam.
Now set that against the construction timeline of the data centers themselves. A hyperscale facility can be built in eighteen to thirty months. A new gas plant takes four to five years. A nuclear plant takes longer still. The load arrives before the light does. This is the structural gap that Entergy's warning encodes โ a mismatch in clocks, where the customer's calendar and the utility's calendar no longer intersect. When those two clocks disagree, the utility has only two honest options: tell the customer to wait, or tell the customer it might be cut off. "Potential disconnections" is the second option, spoken aloud.
I have seen this pattern before, though in a different costume. During the DeFi summer of 2020, I spent months studying how unstable stablecoin pegs rippled through cross-border remittance corridors in Latin America. The mechanism was always the same: a promise of seamless liquidity, backed by reserves that were thinner and slower than the promise implied. When the reserves could not move fast enough, the peg wobbled, and the people at the end of the corridor โ the ones actually sending money home โ paid the price. The structure of the Entergy problem is identical. A promise of abundant, cheap, reliable power was extended to a marquee customer, backed by a generation and transmission system that cannot be built fast enough to honor it. The wobble, when it comes, will be measured not in cents per token but in megawatts of darkness.
The second thing to understand is that the real dispute is not about physics at all. It is about who pays. This is the question the sixty-word brief conspicuously avoided, and it is the only question that actually matters. When a utility adds generation and transmission to serve a giant new load, someone must fund the capital. There are three parties at the table, and their interests diverge sharply. The utility wants to expand its rate base, because under regulated returns a larger asset base can mean a larger profit โ but only if the costs are recoverable. The data center wants cheap, reliable power, and it retains a quiet but potent threat: it can site its next facility in another state. And the residential ratepayers, represented by regulators, have a legitimate fear that they will end up subsidizing the world's richest technology companies, or worse, absorbing what the industry calls stranded costs โ the bill for infrastructure built for a customer who one day leaves.
The weapon in this negotiation is the rate structure, and it has been deployed across the country. In 2024, a wave of large-load tariff proceedings swept through states like Ohio, Georgia, Virginia, and Indiana, as regulators tried to design special contracts that force hyperscale customers to carry their own incremental costs. Georgia Power's large-load applications, dominated by data centers, ran into the multiple gigawatts. Virginia's Dominion faced similar arithmetic. The pattern is unmistakable: the era of welcoming the large load at any price is ending, and the era of pricing the large load's externalities is beginning.
Entergy's warning, read correctly, is a negotiating move. It is the utility signaling to both the regulator and the customer that the status quo is untenable. To the regulator, it says: give me the rate treatment I need, or I cannot serve this load. To the customer, it says: the power you assumed was guaranteed is now a privilege under review. When a utility publicly threatens to disconnect its biggest new client, the probability that the two sides are deadlocked is very high. The threat of darkness is the most expensive bargaining chip in the room.
There is a further layer that the crypto-native reader will recognize immediately, and that the brief's very publication in a crypto venue quietly confirms. The large load is not just an AI phenomenon. It is a crypto phenomenon too. Bitcoin miners were the original grid-scale nomadic customers โ the ones who signed interruptible contracts, chased cheap power, and taught utilities how to think about flexible, mobile, high-density demand. In the years before the AI boom, miners were the load that utilities both courted and resented. Now, as AI data centers crowd into the same queue, the two are being bundled in the minds of regulators as a single category: the large load, the grid stressor, the externality to be managed.
This bundling is not accidental, and it deserves to be named. The choice to publish a utility-regulation story in a cryptocurrency outlet suggests that someone in the editorial chain sees AI compute and crypto mining as the same animal โ two heads of the same demand shock. From a pure engineering standpoint, that framing is defensible. Both consume firm power at scale. Both compete for the same scarce generation and the same congested interconnection queue. Both trigger the same rate-design fights. But from a strategic standpoint, the bundling carries a risk for the crypto industry: when you are categorized alongside the most politically powerful industry on earth, you inherit its enemies as well as its allies.
The enemies, in this case, are forming fast. Several jurisdictions have begun discussing moratoriums on new large loads โ pauses on approvals while regulators catch their breath and redesign cost allocation. The word "moratorium" is the heaviest policy signal in the entire story. It marks the transition from a phase of courting data centers to a phase of managing their side effects. For Bitcoin miners, who have lived through bans and backlashes before, this is familiar terrain. For the AI industry, which has grown accustomed to being welcomed as an economic savior, it is a rude awakening. And for anyone who holds crypto as a bet on decentralized infrastructure, it is a reminder that the scarcest resource in the next cycle may not be blockspace. It may be the electricity required to produce it.
Volatility is the tax on impatience. And the grid, unlike a market, does not clear in seconds. It clears in years โ the years it takes to site, permit, finance, and build a power plant. That mismatch between the speed of digital ambition and the speed of physical infrastructure is the deepest tension in the current boom, and it is the tension the Entergy brief accidentally exposes.
Let me be precise about the scale of the problem, because vagueness is where narratives hide. American data center electricity consumption was roughly 176 terawatt-hours in 2023, about 4.4 percent of national demand. By 2028, the Lawrence Berkeley National Laboratory projects it could reach between 6.7 and 12 percent. The Electric Power Research Institute has floated a figure near 9 percent by 2030. These are not marginal adjustments. They are the equivalent of adding entire new industrial sectors to a grid that spent two decades assuming it would never grow again. And the load is concentrated โ geographically, because it must sit near fiber and water and tax incentives, and temporally, because it arrives in clusters rather than gradually. Concentration is what breaks systems. A grid can absorb diffuse growth; it struggles with a single county that suddenly demands a gigawatt.
This is why the phrase "structural tightness" keeps appearing in my analysis, and why I resist the more comfortable word "shortage." A shortage suggests a temporary imbalance that markets will clear. Structural tightness suggests something harder: a system whose physics, permitting timelines, and supply chains are all moving more slowly than the demand pressing against them. The equipment bottleneck is the clearest evidence. When gas turbines are sold out through 2027 and transformers take three years to arrive, no amount of capital can accelerate the electrons. You cannot financialize your way out of a lead time. That sentence should be printed above the desk of every infrastructure investor, crypto or otherwise.
Now the contrarian angle, because the comfortable reading of this story is incomplete. The obvious conclusion is that the AI boom is walking into a wall of its own electricity demand, and that this is a bearish signal for the entire compute economy, crypto included. I want to resist that conclusion, or at least complicate it, because it mistakes a constraint for a verdict.
The decoupling thesis runs like this. Crypto mining and AI compute are being treated as one demand shock, but their economics diverge in a way that will matter enormously over the next several years. AI data centers need firm, uninterrupted power โ they cannot simply stop when the grid is stressed, because a paused inference is a broken product. Bitcoin miners, by contrast, are structurally interruptible. They can curtail within seconds, sell their power back into the grid at peak prices, and treat volatility itself as a revenue stream. In a world of structural tightness, that flexibility is not a liability. It is the most valuable grid service a large load can offer, and almost no one has priced it correctly yet.
The blind spot in the current panic is the assumption that all large loads are the same burden. They are not. A load that can flex is a stabilizer; a load that cannot is a stressor. As grids tighten, regulators may discover that the very customers they once resented โ the flexible miners โ are the ones who can help them survive the peak. That is a reversal of the conventional narrative, and it is the kind of reversal that markets are slow to recognize and quick to overreact to once they do.
There is a second blind spot, equally important. The rush to serve AI load with new gas plants is quietly undermining the climate commitments of the very companies driving the demand. Amazon, like its peers, has pledged to run on renewable power. But the most likely technology to backstop its new data centers is natural gas, because gas is what can be built fast enough. The result is an uncomfortable paradox: the AI boom may be the single most powerful force pushing fossil generation back onto the grid, precisely at the moment when the decarbonization narrative claimed victory. This is the tension that the ESG frameworks have not yet learned to price, and it is the one I expect to define the next round of institutional scrutiny.
The third blind spot is regulatory fragmentation. American electricity oversight is primarily a state affair, and states are diverging. Some are doubling down on recruitment, offering tax abatements and fast-tracked interconnections. Others are turning cautious, demanding that large loads carry their own costs and pausing approvals while they redesign the rules. This patchwork produces a predictable inefficiency: data centers and miners will migrate toward the permissive jurisdictions, concentrating load and stress in exactly the places least prepared to handle it, while the cautious states watch their economic development hopes drift away. There is no federal framework to reconcile these choices, and the absence of one is itself a market signal โ a signal that the conflict between compute demand and grid capacity will be resolved unevenly, state by state, in a series of small crises rather than one large one.
I find myself returning, as I often do, to the year 2022. During the worst of that bear market, I stepped back from public writing for three months, exhausted by the collapse of leveraged protocols and the human cost that the charts never showed. What I concluded during that solitude was that sustainability in any system โ financial or electrical โ depends on alignment between the promise made and the resources held to honor it. The Entergy brief is a small, sharp illustration of misalignment. A promise of power was extended before the power existed. A customer was courted before the capacity was secured. And now, in the quiet language of a regulatory filing, the bill is being presented.
So what should the reader take from this? Not a trade. Not a token. A frame. The next phase of the digital economy will be decided as much by transformers and gas turbines as by code, and the industries that understand this first will be the ones that survive the transition. Bitcoin mining, whatever its critics say, has been living this reality for years; it is the rare sector that already treats energy as the primary constraint and flexibility as the primary asset. AI, for all its capital and confidence, is only now learning the same lesson. The two are not competitors for a fixed pie so much as fellow students in the same hard school of physical limits.
The brief that prompted this essay was sixty words long, filed under a category that did not quite fit. That misfiling is the tell. The crypto press noticed the story because the crypto world is already entangled in the electricity question โ through miners, through energy markets, through the growing overlap of compute and capital. What looked like an editorial error was in fact an early warning, published in the one venue whose readers might actually understand it.
The lights over the data centers of Mississippi will either come on or they will not. But the deeper question โ who pays for the power, who bears the risk when the promise outruns the plant, and whether the digital economy can grow faster than the grid that feeds it โ will echo long after this particular negotiation is forgotten. Follow the money, not the noise. And follow the load, because in the years ahead, the load will be where the money is made and lost.
The tide does not ask for permission. Neither does physics. The only question is whether we build the capacity before the demand arrives, or whether we learn, once again, that the light we promised was never ours to command.
