A draft Treasury Department report, obtained by NOTUS and not publicly released, argues that an AI downturn would land harder on the institutional investors that underpin the U.S. financial system than the dotcom bust did on speculative retail traders. The document, written by career Treasury analysts and prepared for Treasury Secretary Scott Bessent and Federal Reserve Chair Kevin Warsh, stops short of predicting an imminent crash. It models a path in which investment cuts, lost investor confidence, and slower growth propagate through six named transmission channels: stock markets, private credit markets, companies financing data-center buildouts, cloud providers, chip manufacturers, and utilities.
The most pointed comparison to the late-1990s bust is structural. In 2000, retail investors carried most of the losses. Today, fewer retail dollars are backing AI, and the sector's funding sits with pension funds, insurance companies, banks, and private-credit funds whose solvency matters to the broader economy. According to the draft obtained by NOTUS, that shift is what elevates an AI correction from a market story to a financial-stability story.
The draft does not declare an AI bubble. It frames the risk as conditional on the sector failing to meet stated productivity and profitability expectations. Even on that conditional path, the analysts warn that supply-chain disruptions, geopolitical tensions, electricity bottlenecks, and utilities shortfalls could accelerate the slide, according to NOTUS's reporting on the document.
A Treasury spokesperson told NOTUS the document is unvetted and not representative of the department's policies or views, and that the report awaits final approval. The spokesperson's disavowal does not erase the document. Career civil servants prepared the analysis, named political principals received it, and the document has now surfaced publicly. The political distance between the draft and the department's public posture is itself part of the story.
Each one corresponds to capital flows already visible in earnings, deal flow, and capital-market activity around AI infrastructure. A pullback in any one channel would force markdowns and refinancing decisions across the others, because the same firms increasingly finance one another.
The draft also acknowledges what separates the AI sector from the late-1990s tech cohort. Leading AI firms are, on the whole, more mature, more profitable, and carry stronger balance sheets than the speculative ventures that defined the dotcom era. The Treasury analysts flag this as a meaningful difference, not as a rebuttal of the bubble thesis. Mature incumbents can still over-build, and balance-sheet strength does not insulate a sector whose expected returns depend on continued capex from the same firms that finance each other.
The document sits inside an active Treasury engagement on AI. Treasury's press office recently announced an AI Innovation Series that brings together regulators and industry on AI risks and opportunities. A leaked internal draft on AI bubble risk is not, on its own, a contradiction of that work, but it does show that career staff are conducting parallel analysis that the public-facing program has not surfaced.
Downstream outlets have re-reported the NOTUS findings, with Seeking Alpha and Political Wire among those carrying the same core facts. No contradictory primary reporting has surfaced. The next trigger to watch is whether Treasury releases a finalized version of the report, edits it down, or leaves it in the drawer. A finalized version would carry institutional weight the draft cannot. Until then, the document is one input into a financial-stability conversation that is already happening inside the agencies responsible for managing the next downturn, whether or not the public sees their work.