What the Filings Don't Show: Tracing the Credit Behind the AI Build-Out, and the Disclosure Gaps That Hide It

The money behind the AI boom is real; the disclosure architecture that would let anyone price its risks is not. The agentic-AI credit channel — through which bank- and market-sourced capital finances AI infrastructure — leaves investors, regulators, and counterparties without the records they need to see where the risk actually sits. This article asks a simple question: how much of it can an outsider actually see in the public record? It walks through that record in order — from what is clearly visible, to what is visible only in part, to what no filing reveals because no rule requires it — and treats each gap as a testable claim: if the gap is real, the public record should run out at a specific point traceable to a specific missing rule. The evidence is drawn from Shared National Credit data, the FDIC Quarterly Banking Profile, named-bank 10-K filings, business development company marks, SEC N-PORT position data, depreciation disclosures, GPU-utilisation proxies, buyer-ROI surveys, second-hand GPU price series, and a review of US disclosure rules with an international comparator. From the gaps the article derives a short list of disclosure reforms — each one the direct complement of a gap the public record exposes — and characterises the cost of continued opacity in terms of mispriced systemic exposure.