Entergy's Warning to AWS: Large Loads, Firm Capacity, and the Power Constraint Rewriting Crypto's Macro Case

The item was sixty words long. It ran on a crypto desk — Crypto Briefing — and it described something that, on its chaotic surface, had nothing to do with crypto at all: Entergy Mississippi, a regulated utility, had warned Amazon Web Services that its data centers in the state could face disconnection if load growth kept outrunning the system's ability to serve it. Read it again, slowly, and notice the shape of it. The largest cloud provider on earth — a firm that negotiates fifteen-year power contracts the way other firms negotiate office leases — has been publicly threatened with the loss of electricity. The threat did not come from a regulator, a court, or an activist. It came from the company that sells it the electrons. That a crypto publication thought this worth printing is the first clue. That it is, in fact, the most important energy signal of the quarter — and one of the most under-priced inputs into any crypto thesis — is the second. For most of the last two decades, the macro watcher's instrument panel was simple. Watch the Fed. Watch the ten-year. Watch the dollar. The cost of capital was the gravity that moved every risk asset, and crypto, for all its claims to independence, moved with it — a high-beta proxy for the global liquidity cycle, dressed in the language of sovereignty. That panel is now incomplete. The binding constraint of the late 2020s is not the price of money. It is the price of firm power — electricity available on demand, at the moment it is demanded, seven days a week, at four in the morning, in the middle of a heat wave. The Mississippi item is the sound of that constraint finally biting a customer that assumed it could never be bitten. Set the context. American electricity demand was essentially flat from 2007 to 2022 — a generation of near-zero growth that lulled utilities into a capital-allocation stupor. Then it turned. The EIA and NERC both revised load forecasts upward through 2024. Data centers, on the Department of Energy's own Lawrence Berkeley lab numbers, consumed roughly 176 TWh in 2023 — about 4.4 percent of national demand — a figure LBNL projects could reach 6.7 to 12 percent by 2028. EPRI puts the 2030 number near 9 percent. On top of that sit manufacturing reshoring, electrification of transport and heat, and the quiet compounding of everything the grid was never rebuilt to carry. The supply side did not respond. The interconnection queue — the line of generators waiting to connect — sits at roughly 2,600 GW, against an average wait of about five years. Gas turbine lead times, historically around eighteen months, have stretched into 2027 and 2028; GE Vernova and Siemens Energy are booked well past the horizon most equity models bother to draw. Large power transformers run two to three years. NERC's reliability assessments have grown progressively darker about capacity margins. This is the landscape into which a utility in one of America's cheapest, most power-surplus states is now warning its single largest new customer that the plug might have to be pulled. Mississippi matters precisely because it was, until recently, an exporter of the thing everyone now wants. It is a low-rate, power-surplus state, home to a share of the Grand Gulf nuclear station and a fleet of gas that once had room to spare. AWS's ten-billion-dollar commitment was the state's trophy, the proof that the recruitment strategy worked. And now the trophy is the problem. The margin that made Mississippi attractive is the margin the data center is consuming, and the utility is discovering that the state sold its headroom faster than it can rebuild it. Here is the first thing the crypto analyst must internalize, and it is a reversal of everything the new-energy narrative taught us to expect. The unit of account is no longer the megawatt. It is the firm megawatt — the dispatchable, on-demand, weather-independent unit that can be counted on at 3 a.m. And on that unit, the United States is not oversupplied. It is dangerously short. We spent three years being told the opposite. Solar module capacity is in structural glut; cell and wafer prices collapsed; battery pack prices fell through the floor. But module overcapacity is not power overcapacity. Nameplate capacity and firm capacity are not the same asset, and the market has been pricing them as if they were. A gigawatt of solar producing at a 25 percent capacity factor and delivering nothing at midnight is not a substitute for a gigawatt of combined-cycle gas that delivers at will. When the marginal buyer is a hyperscaler running inference around the clock, intermittency is not a discount. It is a disqualification. There is a clock problem underneath the capacity problem. A hyperscale data center can be designed, permitted, and energized in roughly eighteen to thirty months. A new combined-cycle gas plant takes four to five years; a nuclear unit takes longer than most careers. The load arrives first and the supply arrives second, and the gap between them is a structural window during which the only honest answer to "can you serve me?" is "not yet." That window is open right now across the American South, and Entergy's warning is its local expression. It is not that the utility failed to plan. It is that the planning horizon and the construction horizon have inverted, and no one has reconciled them. This is why the technical roadmap for new American generation is not ambiguous, whatever the press releases say. For the short run — 2025 to 2028 — the answer is gas, and it is gas by an overwhelming margin, north of 80 percent of new dispatchable capacity. For the medium run, existing nuclear life extensions and small modular reactors. Renewables plus storage remain a supplement, not a spine, because firm power is the product the load actually demands. Entergy's warning to AWS is the local photograph of a national decision already made. Which is why the most interesting fight in American power is not between generators but between two ways of organizing supply. On one side sits the traditional model: the regulated utility builds the capacity, spreads the cost across its rate base, and earns a regulated return of roughly nine to ten percent. On the other sits vertical integration — the hyperscaler reaching behind the meter, co-locating with a nuclear plant, signing a direct power purchase agreement that bypasses the utility entirely. The Talen Energy–Amazon co-location at Susquehanna was the test case, and when FERC declined to expand the arrangement in 2024, it drew a line: you may not quietly privatize the grid's firm capacity and leave the public system to absorb the imbalance. Entergy's warning to AWS is the same line, drawn by a different hand. The question of who may own the firm megawatt — the utility, the hyperscaler, or a miner selling its interconnect — is now the central question of the sector. Now bring in crypto, because crypto got here first and nobody gave it credit for it. Long before the hyperscalers started queuing for interconnects, Bitcoin miners had already built the prototype of the large, flexible, interruptible load — a customer that signs for hundreds of megawatts, accepts curtailment on demand, and is paid, in effect, for the privilege of being switched off. In ERCOT, miners became a de facto demand-response fleet, absorbing surplus power in the trough and shedding it at peak, stabilizing a grid that would otherwise have had to build peakers it could not afford. That model is now being repriced, and the repricing is the story. The same physical asset — a substation, an interconnection agreement, a power purchase contract, a shell full of transformers — has been revalued from "hashrate" to "GPU capacity." Miners who spent a decade learning to arbitrage cheap electrons discovered that the hyperscalers would pay them for the interconnect they already owned. The pivot from Bitcoin hosting to AI hosting is not a pivot at all. It is the same asset, the same contract, the same curtailment clause — sold to a richer tenant. And here the industry's favorite story — that this is all decentralized, permissionless, sovereign — meets its cold little mirror. The mining pool that routes the hashrate, the foundation wallet that holds the treasury, the hyperscaler that signs the PPA through a wholly-owned subsidiary: every one of them is traceable, every one of them is a legal person with a registered address, and the decentralization claim functions less as a description of reality than as a compliance shield. The DAO that governs the pool is a Delaware LLC with a lawyer. The "community" that decides the tariff is a committee of three. Trace the wallets, and you find the same names on the power contracts. Which brings us to the black hole at the center of the Mississippi item — the thing the sixty words carefully do not mention. Who pays? The cost of the generation, transmission, and distribution that a data center requires is enormous, and the question of who bears it is a four-way war between the utility, the hyperscaler, the regulator, and the residential ratepayer. The utility wants the load to carry its own incremental cost, because the alternative is a political catastrophe: voters discovering that their bills rose to subsidize a trillion-dollar company's inference cluster. The hyperscaler wants cheap, reliable, fast power, and treats site selection as a credible threat — a gun it can point at any state that asks it to pay its way. The regulator wants growth without the backlash. And the residential ratepayer, who has no seat at the table, wants the lights on and the bill flat, and will vote against whoever fails to deliver both. The mechanism through which this war is fought is the "large load tariff" — a special rate class for customers above some threshold, typically 100 MW, that forces them to fund the infrastructure built on their behalf. In 2024 these tariffs spread across the country: AEP Ohio, Georgia Power, Dominion in Virginia, Indiana Michigan Power. Georgia Power's large-load pipeline ran to roughly 7.6 GW, overwhelmingly data centers. The point of the tariff is to defuse the stranded-cost bomb: if the hyperscaler leaves, the ratepayer should not be left holding the substation. Which is exactly why Entergy's warning is not a technical bulletin. A public threat to disconnect your largest new customer is the nuclear option of a rate negotiation, and utilities do not reach for it unless the talks have stalled. The warning is aimed in two directions at once — at the Mississippi Public Service Commission, which must approve the tariff, and at AWS, which must accept the terms — and it says, in the only language both understand, that the status quo is no longer on the table. Buried in the reporting is a word heavier than any other: moratorium. It signals a phase change in how American states treat the data center. For three years the posture was recruitment — tax abatements, fast-tracked interconnects, low rates, a race to the bottom to land the logo. What the moratorium talk announces is the arrival of the next phase: management of the externality. The state that once begged for the load now reserves the right to cap it, price it, or refuse it. The policy pendulum has crossed the midpoint, and both AI and crypto sit on the far side of it. There is a structural parallel here that crypto readers will recognize instantly, because we have lived it. The industry spent years slicing its own liquidity across dozens of Layer 2 rollups, each one a promise of scale that resolved, in practice, into a fraction of the same small user base — fragmentation dressed as expansion, liquidity bled into ever-thinner pools that shared the same handful of users. American electricity governance is now doing the same thing to the load it is courting. There is no federal framework for large-load interconnection and cost allocation, so fifty states are inventing fifty regimes, and the resulting patchwork is neither efficient nor fair. Fragmentation is not scaling. It is the appearance of it, purchased at the cost of the thing it claims to deliver. The investable consequence is a supply chain that looks nothing like a bubble and everything like a decade-long shortage. Transformers, turbines, high-voltage cable, switchgear: the physical components of a grid that was allowed to age for twenty years are now the scarcest goods in the industrial economy, and their order books extend past the horizon of most equity models. The capacity shortage is not a quarter's anomaly. It is the defining capital cycle of the next ten years, and it is being financed by the same forces — AI compute and flexible crypto load — that the market still discusses as if they were software problems. And now the ethical bill, because there always is one. If Entergy serves AWS's growth the way the roadmap says it will — with new gas — then the cleanest, most profitable customer in the state will have raised its own Scope 2 emissions to get the power it needs. The hyperscalers have pledged 100 percent renewable electricity and net zero; the grid they are connecting to is about to get dirtier, because the firm power is coming from methane. This is the most under-modeled force in the entire climate narrative: AI compute growth may be the single largest driver of fossil capacity additions in the developed world over the next five years, and it is arriving under a green banner. The tension between the climate pledge and the inference cluster is not a rounding error. It is the whole problem, wearing a sustainability report. There is a second-order lesson for Bitcoin specifically, and it is about the difference between subsidy and demand. For years the network's security budget rested on the block subsidy, a declining stream that everyone knew would eventually have to be replaced. Ordinals and inscriptions — whatever one thinks of the cultural circus they produced — injected a genuine, fee-paying demand for blockspace that the network had never had before. Without that demand, the security model's arithmetic would already look broken; with it, the network has proof that a fee market can exist. The parallel to the grid is exact: a system cannot run on nameplate alone. It needs the marginal buyer who shows up and pays, whether that buyer is an inscriber bidding for a block or a hyperscaler bidding for a firm megawatt. Subsidy is a phase. Demand is a structure. Now the contrarian move, the one that destabilizes the tidy reading. The comfortable crypto thesis of the moment is decoupling — the claim that the asset class has finally escaped the macro cycle, that ETF flows and on-chain fundamentals now drive price independently of the Fed. It is a pleasant story, and it is half wrong. Crypto is not decoupling from macro. It is decoupling from the old macro and re-coupling to a new one, whose substrate is physical. Beneath the chaotic surface of the tape, the asset is no longer levered to the discount rate alone. It is levered to firm capacity — to the substation, the interconnect, the curtailment clause, and the tariff. And firm capacity is the scarcest, most politically charged input in the economy. The industry's framing of an "AI versus crypto" war for power is a false binary, and a costly one. Both are large, flexible loads. Both are being repriced by the same tariff mechanism, in the same statehouses, by the same regulators who have learned that the load they once chased now threatens the ratepayer who elects them. The binding constraint on both AI and crypto in the late 2020s will not be the SEC, and it will not be the Fed. It will be the public utility commission. That is the blind spot. Every model in the sell-side stack tracks liquidity and flows; almost none tracks the interconnection queue and the capacity margin, which is where the marginal cost of the next unit of hashrate and the next unit of inference is actually determined. And when capacity binds, the bearish case changes shape. If firm power is the constraint, then the marginal cost of new hashrate acquires a physical floor — a floor set not by hardware cycles but by the five-year wait in the interconnection queue and the two-year lead time on a transformer. The old assumption that mining is an unbreakable race to the bottom meets a wall it cannot argue with: you cannot bid for hashrate that the grid will not energize. So watch Mississippi. Watch the tariff docket, the capacity margin, the color of the next power plant Entergy builds. The sixty-word item that a crypto desk chose to print is not a curiosity; it is a temperature reading, and the temperature is rising. On the chaotic surface, this still looks like a local utility dispute. Underneath, it is the sound of a new constraint being installed at the base of the entire digital economy. The question for the next cycle is not whether crypto decouples from the dollar. It is who holds the bill when the grid says no — and whether the asset class that spent a decade claiming to be independent can survive discovering that it, too, needs the lights on.