Research & Commercial Insight · June 22, 2026 · UAO Research
There is a tidy story large allocators have told themselves about artificial intelligence: we own the winners, we are diversified across the picks-and-shovels, and if the trade corrects we will mark it down and move on, because it sits in the part of the book designed to absorb drawdowns. That story was true in 2024. It is decaying in 2026, because the way the build-out is funded has changed — and the new funding runs straight through the parts of an institutional portfolio that are not designed to absorb a drawdown at all.
Channel one: the migration from equity to credit. For most of the AI cycle, hyperscaler capital expenditure was financed out of operating cash flow and equity — the safest possible funding for a speculative build. That is no longer sufficient to the scale. The BIS, in Bulletin 120, puts hyperscaler bond issuance above $100bn in 2025, with roughly another $120bn routed through special-purpose vehicles and private credit, and private-credit lending to AI-related sectors rising from near zero to more than $200bn. Nvidia's $25bn June deal — its first since 2021, ordered 3.4× over — is the marquee example of a cash-rich firm choosing to term out decades of fixed-rate debt against the build. When capex moves from equity to debt, the risk moves with it: from a sleeve where a 30% drawdown is a bad year to a sleeve where the same disappointment is a default, a downgrade cascade, or a gated vehicle.
Channel two: circularity and the public-private contagion path. The more uncomfortable feature of this financing is how interconnected it is. The same handful of names appear as customers, suppliers, equity sponsors and now lenders to one another; off-balance-sheet vehicles raise private debt against data-center assets that are leased back to the very hyperscalers whose demand underwrites the lease. The Bank of England's Financial Policy Committee has flagged exactly this — AI as more than a third of 2025 private-credit deals, up from about 17%, inside a private-credit market whose "complexity, leverage and interconnectedness could amplify stress." For a universal owner the hazard is specifically that the equity sleeve and the credit sleeve are no longer independent bets on AI; they are the same bet, wearing two different instruments, and they will be marked down together in the scenario that matters.
Channel three: refinancing into a hawkish world. None of this debt is a problem while it can roll cheaply. The regime says it may not. The ECB has hiked, the Bank of Japan has moved to 1.0%, and markets are pricing no Fed cuts in 2026 — what they now read as a synchronized higher-for-longer backdrop arriving just as the first big AI maturities approach. Debt taken on against a multi-decade return assumption has to be serviced and refinanced against a near-term funding cost that is no longer falling. That is the gap the Probability Desk's AI Capex Air-Pocket scenario (currently a 15% reading) is built to stress: not that AI fails, but that returns arrive later than the debt comes due.
So what should a fiduciary actually do? Three disciplines, none of them "sell AI." First, look through the label. Ask each credit and private-credit manager to decompose AI exposure — direct issuer, SPV/lease-back, vendor financing — and to show whether ratings reflect the lease-back leverage or ignore it. Second, price the concentration twice — by name and by input. The same single-country rare-earth dependency that China switched off this morning (see today's brief) is a supply-side air-pocket sitting underneath the demand-side debt. Third, demand disclosure as stewardship, not box-ticking. The universal owner's edge here is that it sits on both sides of the trade: it can use its equity voice to make the borrowers it lends to disclose the very concentration and refinancing risk that threatens its credit. That is the version of "system-level stewardship" that pays its own way — and it is the agenda, conveniently, of the asset owners gathering at PRI Investor Day this week.
The build-out is the largest capital-formation event of the decade, and the universal owner is now financing a growing share of it. The discipline is simple to state: make sure that, when the equity story and the credit story finally diverge, your managers were paid to take that risk — and knew they were taking it.
Sources: BIS Bulletin 120, "Financing the AI boom: from cash flows to debt," 2026; Bank of England Financial Policy Committee (via Central Banking, Reuters), 2026; Bloomberg / SEC Form 424B5 (Nvidia), June 15, 2026; ECB and Bank of Japan policy statements, June 2026; UAO Probability Desk, TR-AI-CAPEX-AIRPOCKET (live).