Washington Renamed AI. The Market Fenced It.

A $500 million collateralized loan obligation issued on September 29 carries a sentence no such document has carried before: a 15 percent ceiling on AI-related debt in the collateral pool. Allstate issued it, anchor investor PGIM pushed for it, BNP Paribas arranged it, and it is the first CLO ever to fence AI exposure in writing ^[1]. The announcement and the cap simply landed on the same day. On September 29 the president launched America.gov and announced he will sign an executive order making "super intelligence" the official federal term, on the reasoning that the word "artificial" undersells it ^[2]. Days earlier, at the United Nations, he had proposed an AI Force and an AI czar and dismissed existential-risk warnings as a hoax ^[3], and by week's end his side was running a meme campaign against "AI doomerism" ^[4]. A document that allocates a loss or moves a price is operative. A document that changes a name is not. Everything operative in this window was written by someone pricing risk. The newest fence in a paper trail running back almost a year. By November 2025, W. R. Berkley, Great American and Chubb were seeking permission to exclude AI-tool liabilities from corporate policies, AIG had filed for generative-AI exclusion options, and Mosaic Insurance was refusing to underwrite large language models outright, on the fear that one model's failure would hit every policy at once ^[5]. Then the spread moved: Apollo's chief economist Torsten Slok says credit default swaps on the hyperscalers have repriced roughly 60 basis points against the banks since October 2025 ^[6].

What the market is repricing is hyperscaler credit fundamentals, namely a debt-financed AI capex cycle with rising leverage, negative free cash flow and uncertain payback on depreciating assets. — Torsten Slok

The state appears in this record only once, and only as a price. The 10-year Treasury peaked above 5.2 percent in late September, and the Fed chair, Kevin Warsh, attributes part of that climb to the buildout ^[7].

The surge in capital expenditures … is real, and the so-called hyperscalers are out in the market raising funding, and so the competition for capital is real and I think it partly explains the increase in yields. — Kevin Warsh

Japan's finance minister says the same thing from across the Pacific, where the two-year yield sits at a 31-year peak ^[8].

we're in a boom, and booms don't last forever. — Katayamazu Onsen

A Kansas City Fed president has gone further, calling the tangle of firms and contracts around data centers a too-big-to-fail ecosystem and reaching for 2008 to say it ^[9]. He can name the comparison but not write the rule. Each fence — the cap, the exclusions — moves risk off its writer's own books and leaves the flow of money alone. Nvidia's finance chief projects top-five hyperscaler capex accelerating to $1.3 trillion by 2027, a year ahead of the old consensus ^[10], and more of it is now debt as the cash-to-capex ratio keeps falling ^[11]. Roughly $4.5 trillion of private-market dry powder stands by, and Goldman's own credit strategist describes markets structuring around AI risk rather than closing to it ^[12]. Every constraint in this record is addressed to somebody's exposure — this pool, this policy, this spread — and nothing is addressed to the flow itself. The $1.3 trillion number sits at the end of it, untouched.