The Plan to Make AI Cheap Broke the Bond Market

The administration's plan for AI was, at bottom, a venture fund's plan. Force the buildout. Make the capital cheap. And, when the capital refused to stay cheap, capture the central bank that sets its price. Each leg was pursued in the open; the third was the one that mattered, and it is the one that has now failed in the most expensive way available. Trump said the quiet part out loud about the financing. His strategy was to borrow short and wait out the Fed chair.

What I'm going to do is I'm going to go very short-term. — Donald Trump

Wait until this guy [Fed Chair Jerome Powell] gets out, get the rates way down, and then go long-term. — Donald Trump

Borrow short now, force rates down later, then lock in long. The whole thing assumed a central bank that would eventually cooperate. The first leg worked. The buildout was forced through every lever available: the Defense Production Act, the EPA reinterpreting the Clean Air Act so construction could begin before air permits cleared ^[1], a pact requiring tech companies to fund new power plants in the PJM region ^[2]. Whatever the local fights, the federal machinery delivered. The plants are being built. The second leg is where the plan ate itself. The buildout's own demand is the inflation. The Minneapolis Fed found AI hardware demand added roughly 0.4 percentage points to core inflation by July — about the same as the tariffs ^[3]. The Fed's own minutes named AI and tariffs as the primary inflation drivers ^[4], and Governor Lisa Cook and New York Fed President John Williams pointed to the $1.5 trillion in announced data center plans pushing up memory chips, electricity, and copper ^[5]. The borrowing to build is also the borrowing that crowds the bond market: the ECB warned US tech giants' AI bond issuance could reach $1 trillion by 2028 and raise financing costs for everyone, governments included ^[6], and the trillion-dollar funding gap Apollo has been warning about ^[7] is the same money the Treasury now has to compete with. Cheap capital was the second requirement, and the buildout itself priced it out of existence. So the administration reached for the third leg: the central bank. Stephen Miran, a Fed board nominee, proposed a "third mandate" requiring the Fed to moderate long-term rates ^[8]. Trump pressured Powell to cut "by a lot" and floated a shadow chair to undercut him ^[9]. Then he nominated Kevin Warsh to replace Powell — a man markets read from the start as a hawk, the kind who would shrink the balance sheet and let yields rise ^[10]. He asserted his independence, and then he did something stranger. He killed forward guidance.

We're going to deliver price stability in the U.S., that's what this committee has signed up to do, and our objective is to do that. — Kevin Warsh

The market read silence as inaction. The result was what analysts called a credibility shock — an 840-point Dow drop and three FOMC dissents in favor of a hike, the first early-tenure dissension in 56 years ^[11]. JPMorgan moved its rate-hike forecast to December 2026 after calling Warsh's press conference "the most troubling since 2012" ^[12]. The central bank meant to calm the bond market produced the opposite. The Treasury ran the same play and got the same result. Bessent expanded buybacks and threatened the traders who kept selling.

Bet against me if you want. — Scott Bessent

The 30-year yield stayed above 5% for 27 straight days, the longest stretch since 2007, and global investors began a "Sell America" trade ^[13]. One leg held, one leg broke, and the third made the whole structure worse than if it had never been tried. The bond market is less stable now than before the capture attempt — and the man installed to cut rates is the reason.