The bond market is not tight because Washington is profligate. It is tight because Alphabet, Microsoft, Meta, and Amazon are hoovering every long-dated dollar on Earth to pour concrete, copper, and cooling towers for AI, and the Treasury is the only seller with the duration to meet them. The diagnosis is not the deficit. The diagnosis is duration.

Here is the one beat of concession I will grant the deficit hawks: yes, Treasury has to roll a wall. Yes, the auction calendar is heavy at the 10- and 30-year. Yes, dealer balance sheets are not what they were in 2010. And yes — the deficit itself is larger than any honest benchmark of full-employment countercyclical policy can defend. An agenda that does not own that fact is not serious. But owning it does not mean bargaining it against the long bond. The interest bill is a consequence of the rate path, not its cause. Tightening the mandatory outlays to “save the long end” is a confidence trick played on the people who hold the long end and on the people who would benefit from those outlays. Cutting entitlements does not flatter the curve. It shrinks the tax base that pays the coupons.

The receipts for the long-end thinness are in the public record. In the quarters before the buyback program’s expansion, the primary-dealer allotment share on long-end auctions had drifted well above historical norms. Tails widened and bid-to-cover ratios thinned in the same window. The Treasury Borrowing Advisory Committee itself, in the minutes covering that period, flagged the long end as a market where dealer balance-sheet capacity had become a binding constraint on auction absorption. A 20-year reopening stopped out at a 3 basis-point tail with a 2.05 bid-to-cover — the kind of auction that prints receipts on the buyer’s appetite. The buyback was the technical response: a seller of duration meeting a market where the natural buyers had rotated out. It is what a Treasury that funds in thirty different buckets does when one bucket leaks. The piece in question mistook the wrench for the sabotage. The wrench is not the cause of the leak. The leak is that patient capital is being allocated elsewhere — into compute, into power, into the equity of the labs building the models. The deficit is the column inch. The capex is the column.

Look at the longer receipts. In 2010, with a federal deficit of $1.3 trillion and a 10-year yield that peaked around 3.3 percent during the Greek scare, the long end barely flinched. In 2020, with a $3.1 trillion deficit and the 10-year collapsing toward 0.5 percent during the COVID panic, the long end rallied. The deficit roughly tripled and yields fell by 280 basis points at the peak-to-trough comparison. Deficits do not print yields. Buyers do. The same nominal deficit stock sat in two entirely different yield regimes depending on who was on the other side of the trade. In 2010 the foreign bid was hungry. In 2020 the cohort with the longest-dated liability on its books was pensions and insurers fleeing deflation. Today that cohort is no longer fleeing deflation. It is sitting on the most enormous private capex cycle in the history of capitalism, and that capex is denominated in 30-year concrete, 50-year power purchase agreements, and 40-year fiber indefeasible rights of use. A single AI campus draws power-equivalent demand that pencils at thirty-year amortization assumptions. The marginal buyer of duration is now a hyperscaler hedging a capex schedule that runs to 2030 and beyond. The marginal seller is the U.S. Treasury trying to fund the same horizon.

Term premium is not exogenous. Term premium is set by the scarcity of long-dated savings relative to the demand for long-dated liabilities, and when hyperscalers are committing tens of billions to data centers whose useful life runs thirty years, the demand for duration at that horizon spikes. The fix for term premium is not fewer entitlements. The fix for term premium is more patient capital, deployed at the same duration the capex demands. And the deficit hawks are selling you austerity because austerity is what their donors wanted to sell all along — they just needed a bond market to dress it up in.

This is the part the editorial pages cannot say. Because to say it requires admitting that AI capex is large enough to move a $27 trillion Treasury market, and admitting that means admitting that the productive structure of the U.S. economy has rotated, in eight quarters, into an infrastructure cycle whose scale the bond market has never seen. And that means admitting that the bond vigilantes are not punishing Washington for being profligate. They are pricing an AI capex absorption that the political class has decided not to name, because naming it would require doing something about it. Doing something about it would require admitting that the policy framework of the last forty years — starve the public sector, supply-side, deficits-are-always-bad — does not have a tool for the problem. They would rather blame entitlements than build a tool.

So build the tool. Four of them.

A public infrastructure bank. Not a PPP shop, not a loan-guarantee factory. A real balance-sheet institution modeled on the Reconstruction Finance Corporation, the European Investment Bank, and the Tennessee Valley Authority. Capitalize it with a one-time Treasury equity injection — call it $250 billion — ringfenced for AI infrastructure: data centers, transmission, cooling, fiber, the physical substrate of the capex cycle that is currently being financed by hyperscaler balance sheets and is therefore being socialized by them. Then give the bank the authority to issue its own long-dated paper — 30-, 40-, 50-year maturities — at a Treasury-rate spread, and on-lend that paper directly into AI infrastructure projects whose cash flows match the duration. The duration matches the capex. The funding leg is sovereign. The risk sits on a public balance sheet designed to hold it, not on a dealer balance sheet that cannot. This is not a subsidy. It is a duration match between the public balance sheet and the public infrastructure.

A public-option compute utility. A federally chartered co-op or public utility standing up compute and data-center capacity at wholesale, with member-ownership open to any household, worker cooperative, or small business. The AI labs already know that their bottleneck is not chips; it is power, cooling, and the time-to-market for a new site. A public utility that takes siting, permitting, and grid integration out of the speculative real-estate market is not “industrial policy.” It is a co-op. It is the same institutional form that brought rural electrification in forty years. The default customer is the small operator who currently rents compute from one of two firms and lives at their mercy.

A sovereign wealth fund democratized through worker and retiree participation. Alaska does this with oil. Norway does this with North Sea receipts — a two-trillion-dollar fund that owns about 1.5 percent of every public company on earth and mails the dividend to the future. There is no honest reason the rents from federally-supported AI infrastructure — the equity stakes the public is already being asked to underwrite through CHIPS Act grants and DOE loan guarantees — cannot flow into a fund whose participants are workers and retirees, not dynastic foundations. Mandate the structure. Cap individual account sizes. Let the dividends compound.

Mandated cooperative ownership stakes in federally-supported AI infrastructure. If the federal government is taking the construction risk — through loan guarantees, EDA grants, the Defense Production Act authorities the AI labs are quietly leaning on — then the public is the equity. The deal should be written down. Worker cooperatives, community land trusts, and employee stock-ownership plans should be the default ownership vehicle for the operating layer of any federally-supported data-center campus. This is not a tax. It is a return on risk capital the public has already put up.

These are not technocratic tweaks. They are institutions. They take the duration mismatch off the long end by giving the public a direct claim on the buildout, and they give the long-end investor a counterparty the deficit-hawk script never imagined: the public, at scale, with a thirty-year horizon of its own.

The deficit column will get its turn. The mandatory outlays will get their turn. Defense will get its turn — and when defense does, the conversation will not be about trimming it to flatter the curve, because the curve is not listening to the column inch. The curve is listening to where the capex is going. So build the tool. The fights over the column inch can come later. First decide who owns the future, before the coupons come due.