The Quiet Exit: Who Ends Up Holding AI Credit Risk

The banks that originated the largest AI infrastructure debt packages have started passing the risk on. Quietly, structurally, and at scale. An analysis of who ends up holding the paper when informed originators exit.

In our July market note we asked a question most allocators had not yet asked themselves: who actually holds the paper behind the AI capex cycle? Over the past weeks we went one layer deeper, tracking syndication data, risk-transfer activity, and regulatory publications. The answer is now observable in the market. The banks that originated the debt have started passing it on. Quietly, structurally, and at scale.

The Largest Construction Loan in History Is Looking for New Owners

In late 2025, JPMorgan and MUFG led a roughly $38 billion debt package to finance Oracle-leased data centers in Texas ($23.25bn) and Wisconsin ($14.75bn), built by Vantage and ultimately serving OpenAI workloads. It was the largest data-center financing ever completed, priced around 250 basis points over benchmark with a four-year maturity.

What matters is not the size. It is what happened next. The lead banks have spent more than six months distributing that debt, and portions have reportedly been offered to non-bank lenders at a discount. When originators sell below par within months of closing, that is a price signal. The institutions with the best information about these loans are willing to pay to reduce their exposure to them.

The Mechanism: Risk Leaves, the Loans Stay

The preferred instrument is the significant risk transfer, or SRT. A bank keeps the loans on its balance sheet but buys default protection on the first-loss tranche from outside investors, releasing regulatory capital. The SRT market is no longer niche: outstanding SRT-linked loan pools reached roughly €800 billion by end-2024, and issuance keeps setting records. The Basel Committee dedicated a paper to the structure in February 2026.

Data-center SRTs are the newest and fastest-growing slice of that market. Morgan Stanley began exploring an SRT on its AI infrastructure loan book in December 2025. RBC followed in June 2026, weighing a transfer tied to roughly $2 billion of data-center-linked financing. In May 2026, the Financial Times reported that global lenders were pursuing private deals and risk transfers specifically to avoid, in the words of one participant, "choking" on data-center debt.

Who takes the other side? Private credit funds, hedge funds, insurers, and pension plans. The mezzanine tranches, where the real risk sits, are concentrated among a small group of specialist funds, some of which buy with borrowed money. Risk does not disappear in this system. It migrates to balance sheets with less regulation, less transparency, and less capacity to absorb a correlated shock. At the end of the chain, once again, sits retirement savings.

Why This Matters More Than the Headline Numbers

AI-related debt issuance is heading toward $570 billion for 2026. Debt tied to AI is already the largest single segment of the US investment-grade market. Roughly $800 billion more sits in private credit and off-balance-sheet structures where mark-to-market discipline is optional. Our July note described the reflexive loop financing the buildout. This is the next stage of that loop, and it carries three implications.

First, the marginal holder of AI credit risk is no longer the underwriter. The institutions that structured these loans, that ran the diligence and know the collateral, are reducing exposure. The investors replacing them are buying yield, not information. Historically, that handoff from informed originators to yield-seeking holders marks the late phase of every credit cycle. It was the defining signature of 2006 and 2007.

Second, the transfer is synthetic, which means the risk can come back. SRT protection has to be rolled. If credit conditions tighten, if the specialist funds face redemptions, or if leverage behind the mezzanine buyers unwinds, new protection becomes unavailable or unaffordable, and the risk lands back on bank balance sheets at precisely the moment banks are least able to carry it. A mechanism marketed as risk dispersion becomes, under stress, a risk boomerang.

Third, the discount is the tell. Public equity markets still price the AI complex for perfection. Ten stocks are roughly 43 percent of the S&P 500. Yet in the loan market, the same trade already changes hands below par. Two prices for one risk cannot both be right. Somebody in this market is wrong, and it is rarely the seller with the information advantage.

The Portfolio Consequence

We are not forecasting the moment this repricing reaches public markets. Reflexive cycles are unforecastable by construction. But the sequence is familiar: risk migrates quietly, spreads stay tight, indices stay calm, and then a single re-labeling event forces every holder to reprice at once. Equity concentration, credit spreads, and growth expectations would move together, because they are the same bet held in different wrappers.

That is the environment the Sigma Horizon Program is built for. Structural convexity held as a permanent position, carried positively, so that waiting costs nothing. Risk-off portable alpha designed to be uncorrelated to the equity complex rather than merely diversified against it. And a dedicated satellite on the Nasdaq, the index where the concentration is most extreme. Convexity is cheap while the story holds and unavailable at any price once it does not. The banks, it appears, have started buying theirs.

Sources: Financial Times (May 2026); Bloomberg (September 2025, October 2025, June 2026); Fortune (December 2025); Basel Committee on Banking Supervision (February 2026); BIS Quarterly Review (March 2026); Risk.net; Forbes (July 2026). Figures as reported by public sources and subject to revision.

Photo: Aerial view of data centers in Ashburn, Virginia (Loudoun County). Credit: Theodore Christopher. License: CC0 / Public Domain.

This publication is issued by Black Flower Capital GmbH (Zug, Switzerland) and Black Flower Capital Management SARL (Luxembourg). It is intended for professional and institutional investors, is provided for information purposes only, and does not constitute investment advice, an offer, or a solicitation. Past performance is not indicative of future results.