The AI Buildout's Borrowing Has Reached the Household Balance Sheet

PIMCO found something odd this summer. The big AI spenders sell bonds in two currencies, dollars and euros: same company, same credit, same promise to repay. Yet their dollar bonds trade more violently than their own euro bonds ^[1]. That is the tell. If the borrower were weakening, both would wobble. The wobble sits in only one channel: the U.S. dollar market, which has swallowed too much of this paper too quickly. A pool of money doesn't refuse a borrower outright. It starts charging more, or trading jumpy, and the borrowing walks to the next pool. That is the whole arc of AI financing this year, and the pools are emptying in sequence. The dollar market went first. Goldman Sachs counts nearly $600 billion of AI-linked debt supply in 2026 ^[2], on the way to roughly $4.1 trillion through 2030 ^[3]. In late September, junk-bond risk premiums touched five-month highs ^[2]. Goldman's analysts have a name for the feeling.

We’re probably past the peak of supply in 2026 for the AI related theme because there’s been a bit of indigestion and fatigue and I think 2027 will be the accelerator. — Goldman Sachs Group, Inc. v. Arkansas Teacher Retirement System

Not everyone feels it equally. Amazon, Google and Microsoft still borrow cheaply; Morgan Stanley's Hugh Briscoe notes these remain solid credits paying little for their ratings ^[4]. The strain hits further down: SoftBank just paid 9.75% on junk bonds to secure AI capacity, and CoreWeave disclosed that a one-point rise in rates adds $30 million to its interest bill ^[3]. This is not a general market seizure either. In mid-August, U.S. financial conditions were the loosest since 1996 ^[5]. The fatigue is specific to where the AI paper piles up. So the borrowing moved. Amazon ran a $25 billion European offering; Alphabet made first-time bond sales in sterling, francs and euros ^[4]. The issuers say it themselves: dollars are now the expensive place to raise money ^[4]. The European Central Bank has noticed the traffic. U.S. tech is close to 10% of new euro-zone issuance, and the ECB warns it may crowd out ordinary European borrowers ^[6]. Central bankers have begun naming the thing. Lagarde called AI's financing needs a primary driver of global yields.

It's not a Euro specific issue. — Christine Lagarde

Japan's finance minister pointed the same direction from the far side of the world, blaming global forces — the AI boom among them — for her country's 30-year-high bond yields ^[7]. Money that used to stay home now chases one buildout across borders. The newest pool is the household, and it was wired in last week. Robinhood switched on AI trading agents for all 29 million of its customer accounts, agents already making close to 30 million transactions a day, with a coming feature that lets them keep trading while the owner is logged out ^[8]. The company is also repositioning itself to hold household savings on its platform for the long term, not just move them through ^[9]. The ambition is stated plainly.

The idea behind agentic trading…[is] every capability a human can do will be available to an AI agent. — Vladimir Tenev

The numbers behind the push: a rival broker's U.S. chief expects AI agents to carry 20% of platform trading volume by year-end ^[10], and a Morgan Stanley survey found 75% of retail investors want exposure to private firms ^[11]. One fact should give anyone pause. In mid-September, individual-investor sentiment hit a 16-month low, with more than half bearish and most sitting in cash ^[12]. The pool is being wired while its owners are scared. The oversight turn arrived in the same fortnight. Former SEC chair Gary Gensler warned of the systemic risk of firms adopting identical AI strategies, and of agents able to move depositors' cash fast enough to crack a fragile bank ^[13]. Kansas City Fed president Jeff Schmid asked the question on the record.

Where we have to start to really synthesize what's happening in the AI and the data center build-out is are we moving to a too-big-to-fail AI ecosystem. — Jeffrey Schmid

Too big to fail, in plain terms: a web of firms and contracts so entangled that if it wobbles, the public has to catch it ^[14]. It is a 2007 kind of question, asked in the same week the last pool of money got wired in.