Universal Owner Risk Radar — the AI build-out is being financed in the dark, and the long-horizon allocator is on both sides of the trade
The defining capital story of 2026 is not how much the hyperscalers are spending — it is who is lending them the money. Amazon, Microsoft, Meta, Alphabet and Apple invested roughly $350 billion in data centres and supporting infrastructure in 2025 and have signalled close to $650 billion more in 2026. Public balance sheets cannot absorb that alone, and increasingly they are not trying to. The marginal dollar of the AI build-out is being raised in private credit and bespoke, off-balance-sheet structures — and the buyers of that paper are the same pensions, insurers and sovereign funds that define the universal-owner audience. For a fund that effectively owns a slice of the entire economy, this is the rare risk that cannot be diversified away, because the fund is simultaneously the lender, the equity holder, and the systemic backstop.
The scale, and where it sits. Technology-sector bond issuance passed $200 billion in 2025, and JPMorgan has estimated AI contenders may need to raise as much as $1.5 trillion in investment-grade bonds over five years. But the headline public deals understate the shift, because the fastest-growing channel is private. Meta's Louisiana build was financed through a roughly $27–29 billion structure led by Blue Owl and PIMCO; Apollo deployed more than $40 billion into next-generation data centres and infrastructure in 2025 alone. The private-credit managers — Blackstone, Blue Owl, Apollo, PIMCO, KKR, BlackRock — have become the primary external financiers of compute. The appeal to an allocator is obvious: long-dated, contracted, inflation-aware cash flows secured against physical assets, at a yield premium to public credit. The problem is that the premium is partly compensation for opacity.
What the regulators see that the rate sheet doesn't. The IMF's April 2026 Global Financial Stability Report devoted its second chapter to private credit, and its language is unusually direct for the Fund: relatively fragile borrowers, a growing share of semi-liquid vehicles, multiple layers of hidden leverage, stale and potentially subjective mark-to-model valuations, and unclear interconnections between participants. The Financial Stability Board's May 6, 2026 Report on Vulnerabilities in Private Credit reaches the same conclusions from the macroprudential side, flagging in particular the migration of risk into insurance and pension vehicles — many now controlled by private-equity sponsors — where valuation discretion and liquidity mismatch compound. Both bodies are careful to say systemic risk is, for now, contained. Both are equally careful to say that this depends on the asset class not continuing to grow exponentially under limited oversight — which is precisely what the AI financing wave guarantees it will do.
Why this lands differently for a universal owner. A stock-picker can underweight data-centre credit and move on. A universal owner cannot, for three reasons. First, exposure is already structural: institutional inflows into private credit ran near $300 billion in 2025 (Mercer), and the asset class sits inside the alternatives sleeve that has grown from a rounding error to roughly a quarter of large pension portfolios. Second, the valuations are reflexive: because so much private-credit paper is marked to model, a fund's reported risk looks lowest exactly when the underlying concentration is highest — the AI build-out is a single thematic bet wearing the clothing of diversified income. Third, the owner is the backstop: if a financing air-pocket forces a wave of mark-downs, the same pensions and insurers absorb it through their own books and through the public markets they own in aggregate. The academic literature is catching up to the practitioners here — a 2026 working-paper synthesis of private-credit markets (arXiv) maps exactly this tension between the asset class's genuine economic function and its fragility to a common shock, and notes how thin the disclosure is for a market this systemically large.
The build-out is real — the question is the financing's resilience. None of this is an argument that the AI capital cycle is a bubble, or that private credit is illegitimate. The compute is being built, the contracts are real, and the long-horizon owner has every reason to want exposure to the physical infrastructure of the next economy. The argument is narrower and harder: the universal owner should price the financing structure as carefully as the asset, because the two have been quietly decoupled. That means demanding look-through to the leverage stacked above the assets a fund lends against; treating model-marked private credit as the concentration it is rather than the diversification it reports as; and asking, before adding to the sleeve, whether the yield premium is paying for genuine illiquidity or merely for the absence of a price. The owners financing the AI build-out are doing the economy's work. They should make sure they are also the ones who know what they hold.
UAO Research. Analytical view, not investment advice. AI-related disclosure per house policy: researched and edited by the UAO editorial desk.
Sources
- Global Financial Stability Report, Chapter 2 — The Rise and Risks of Private Credit, International Monetary Fund, April 2026.
- Report on Vulnerabilities in Private Credit, Financial Stability Board, 6 May 2026.
- Private Credit Markets: Theory, Evidence, and Emerging Frontiers, working paper, arXiv, 2026.
- PIMCO and Blue Owl to lead ~$29bn financing for Meta's AI data-centre expansion, Private Equity Insights, 2025–26.
- Why pension funds are doubling down on private credit despite deepening cracks, CNBC, May 8, 2026.
Continue the briefing. Read the daily brief · watch the daily video briefing · listen to The Universal Owner · view the chart of the day.
Produced and edited by the UAO editorial desk. Not investment advice.