Pro
Pro

The Collateral Question: What the AI Data-Center Financing Boom Means for Private Credit’s Wealth-Channel Investors

The Collateral Question:

What the AI Data-Center Financing Boom Means for Private Credit's Wealth-Channel Investors

Global data-center investment is on pace to reach roughly $2.9 trillion through 2028, and Big Tech’s own operating cash flow is expected to cover only about $1.4 trillion of it, according to Morgan Stanley research — leaving a financing gap of roughly $1.5 trillion that credit markets, not equity, are being asked to bridge. Morgan Stanley projects that private credit will supply the largest share of that bridge, roughly $800 billion, with a further $200 billion coming from corporate bond issuance and $150 billion from securitized products. Total AI-related debt outstanding is on track to approach $570 billion in 2026 alone, and the marginal dollar of financing is shifting away from public bond markets and toward private credit and off-balance-sheet vehicles as coverage ratios on hyperscaler bond deals have fallen from roughly five times in February to under two times by July.

The scale of the number matters less than where it sits. A large share of this lending is concentrated in a small number of borrowers — a handful of frontier AI model developers and hyperscale cloud platforms — financed increasingly through structures that use the underlying chips themselves as collateral. Facilities are built to last decades; the GPUs inside them are generally depreciated over about six years for accounting purposes, while some engineers and project-finance lawyers put their genuinely useful life closer to three or four years, and rental rates for widely used AI chips have already fallen 70 to 90 percent since 2023. That mismatch, more than the headline size of the financing, is what is now drawing sustained attention from regulators, rating agencies, and litigators alike.

That attention is no longer theoretical. A draft U.S. Treasury report, disclosed in early July, compared aspects of the AI investment cycle to the dot-com bubble and warned that a downturn would ripple through stock markets, private credit markets, data-center financiers, cloud providers, chipmakers, and utilities alike. Separately, U.S. insurance regulators voted on June 23 to overhaul how insurers must hold capital against collateralized loan obligations and other collateral-backed loans — precisely the structures increasingly used to finance this build-out — while law firms are already mapping litigation exposure explicitly modeled on the mortgage-backed securities disputes that followed 2008, in case underwriting assumptions prove wrong.

None of this is a reason to exit private credit, an asset class that remains, on the evidence to date, one of the more durable growth stories in the wealth channel. It is a reason for the managers, private banks, broker-dealers, and advisors who have spent the past several years placing client capital into interval funds, non-traded business development companies, and asset-backed lending strategies to ask a more specific question than they have generally been asking: not simply what the coupon is, but what stands behind it, how concentrated it has become, and what happens to the collateral if the assumptions underneath the AI buildout do not hold.

1. The Financing Bridge: How Private Credit Became AI’s Marginal Lender

The arithmetic behind the data-center boom has become a familiar reference point in institutional research: roughly $2.9 trillion in global data-center capital expenditure is expected through 2028, against which the hyperscalers’ own operating cash flow is projected to cover only about $1.4 trillion. The remaining $1.5 trillion must come from somewhere, and Morgan Stanley’s own bridge for that gap allocates the largest share — about $800 billion — to private credit, with roughly $200 billion from corporate debt issuance and $150 billion from securitized products layered on top. Annual data-center securitization issuance, which ran near $27 billion in 2025, is projected by JPMorgan to reach $30 billion to $40 billion in both 2026 and 2027, a rising 7 to 10 percent slice of combined asset-backed and commercial mortgage-backed issuance.

What has changed since the start of the year is the willingness of public bond investors to keep absorbing this financing at the pace the buildout requires. Coverage ratios on hyperscaler bond offerings — a rough gauge of how many dollars of orders show up for every dollar on offer — have fallen from close to five times in February to under two times by mid-July, a sign that fixed-income investors are beginning to price in the size and duration of the commitment being asked of them. As that appetite cools, more of the incremental financing is migrating into vehicles with less public disclosure: private credit funds, direct-lending platforms, and off-balance-sheet special purpose vehicles built specifically to hold datacenter and GPU-backed debt. Several of the largest private-credit sponsors have built dedicated infrastructure and asset-backed lending platforms explicitly to compete for this business, a rational response to genuine demand — but one that concentrates a fast-growing category of risk inside exactly the fund structures that have been distributed most aggressively to the wealth channel over the past three years.

2. Concentration: A Small Number of Borrowers, a Very Large Number of Lenders

The borrower base behind this financing wave is unusually narrow. A handful of frontier AI model developers and hyperscale cloud operators account for a disproportionate share of the demand for data-center capacity, and therefore for the debt raised to build it. Insurers, drawn to the long-duration profile of this lending as a natural match for their own long-dated liabilities, have become significant buyers of it; insurance brokers including Gallagher have already flagged that individual projects concentrating $10 billion to $20 billion of assets in a single physical location are straining underwriting and reinsurance capacity in ways the market has not previously had to absorb at this scale.

The credit quality of what insurers currently hold looks reassuring on its face. U.S. insurers held roughly $276.8 billion in collateralized loan obligations at year-end 2024, up about 2 percent from the prior year, with roughly 80 percent rated investment grade or higher, including about 39 percent rated AAA. What regulators are focused on is not today’s rating stack but tomorrow’s: whether concentration in a small set of borrowers, facilities, and geographies, layered inside increasingly complex structures, is being priced and capitalized correctly for a category of collateral that did not exist at this scale even three years ago.

3. The Collateral Mismatch: GPUs Age Faster Than the Debt Backing Them

Data centers are built to operate for decades. The chips inside them are not. Companies typically depreciate GPUs over about six years for accounting purposes, but project-finance lawyers and engineers involved in structuring these deals generally estimate a genuinely useful life closer to three or four years — and some analysts believe the real figure, as chip generations turn over faster, may be shorter still. Rental rates for widely deployed AI chips have already fallen 70 to 90 percent since 2023, a preview of how quickly the economics behind a facility’s revenue assumptions can shift well before the debt against it matures.

The market’s comfort with this collateral is nonetheless growing, not shrinking: one cloud-compute provider closed an $8.5 billion, investment-grade-rated loan secured directly against its GPU fleet earlier this year, a structure that would have been unusual even eighteen months ago. Layered against that comfort is a stark set of underlying numbers: the AI sector generated an estimated $60 billion in revenue in 2025 against roughly $400 billion in capital expenditure, with an additional $1.5 trillion in external financing estimated to be needed by 2028. Cross-default provisions embedded in most data-center loan agreements mean that a shortfall at one facility, or one borrower, does not necessarily stay contained to that facility or borrower — it is the arithmetic, and the interconnectedness, more than any single number, that lenders and their downstream fund investors now need to underwrite.

4.Regulators and Rating Agencies Start Asking the Same Questions

A draft report circulating inside the U.S. Treasury Department, whose existence was first disclosed in early July, compared aspects of the AI investment cycle to the dot-com bubble of the early 2000s and concluded that AI companies are now more deeply embedded in the broader economy than their late-1990s predecessors were. The report’s authors found that a downturn in the sector would be felt well beyond technology stocks — across private credit markets, the companies financing data-center buildouts, cloud providers, chip manufacturers, and utilities — slowing growth broadly even if it stopped short of a dot-com-style crash.

Insurance regulators moved in parallel, not in response. On June 23, the working group responsible for insurers’ risk-based capital rules adopted new capital factors for collateralized loan obligations, effective with year-end 2026 reporting, and separately overhauled how collateral loans are charged for capital purposes — replacing a flat 6.8 percent charge that applied regardless of the underlying collateral with a framework tied to what actually backs the loan, effective for year-end 2027. Those changes sit alongside a broader tightening already underway: a more principles-based bond definition, annual stress testing for CLOs, a 45 percent capital charge on CLO residual tranches, and closer scrutiny of both asset-adequacy testing and offshore, asset-intensive reinsurance arrangements. None of this is being described by regulators as a response to a crisis already underway. It reads instead as plumbing being reinforced ahead of one, on the reasonable premise that the capital rules written for a smaller, more conventional private-credit market may not fully capture the risk profile of a market now financing purpose-built, single-tenant, technologically fast-depreciating infrastructure at unprecedented scale.

5. The Litigation Echo: What 2008’s Mortgage Cases Suggest About 2026–28

Structured-finance litigators have begun mapping this cycle against the last one. If losses materialize inside the special purpose vehicles used to warehouse and securitize data-center debt, the most likely legal exposure runs through breach-of-representations-and-warranties claims brought by SPV investors against the sponsors who structured the deals — the same theory that underpinned the post-2008 residential mortgage-backed securities “putback” litigation, in which trustees ultimately recovered more than $36 billion from sponsors accused of misrepresenting the assets they had transferred into securitization vehicles.

The parallel is not exact, and no one is predicting a repeat of 2008. But the structural features that made those disputes so costly — layered, off-balance-sheet vehicles; cross-default provisions capable of transmitting a single facility’s problems across an entire financing structure; and limited transparency into what, exactly, sits inside any given pool — are present again here, at a scale legal advisors describe as among the largest privately financed infrastructure build-outs in memory. For fund sponsors and their investors, the lesson is less about predicting a default than about building the documentation, disclosure, and underwriting record now that would be needed to defend a deal’s structuring years from now, if it is ever tested.

6. What This Means for Managers, Advisors, and the Wealth Channel Across the Americas

Many of the vehicles carrying this exposure are the same interval funds, non-traded business development companies, and tender-offer structures that have anchored the wealth channel’s push into private credit over the past several years — the subject of considerable attention in this publication’s own recent coverage of redemption activity across those same fund types. What is different now is the layer sitting underneath the coupon: fact sheets and marketing materials that describe a strategy as “infrastructure debt,” “specialty finance,” or “asset-backed lending” do not always make clear how much of that exposure traces back to a small number of AI borrowers, facilities, or geographies.

For asset managers, the differentiator over the next several quarters is likely to be look-through disclosure: the ability to show, at the borrower, sector, and facility level, exactly how concentrated a fund’s data-center and AI-adjacent exposure has become, rather than relying on broad asset-class labels. For private banks, broker-dealers, and independent advisors — including those serving Latin American and U.S. offshore clients who often reach these strategies through feeder funds, discretionary mandates, or model portfolios — the due-diligence question needs to move beyond yield and toward collateral: what backs the loan, how fast does it depreciate, how many other loans share its fate through cross-default provisions, and what capital or liquidity buffer exists if utilization assumptions come in below plan. None of this argues for treating AI infrastructure financing as a bubble to avoid; it may yet prove to be one of the structural growth opportunities of the decade. It does argue for making sure the diligence embedded in prospectuses, marketing decks, and advisor conversations has caught up to the size of the bet already placed.

Sources

Disclaimer:

  1. This white paper is produced by LYNK Markets for informational and educational purposes only. It does not constitute investment, legal, tax, or financial advice, or a recommendation of any security or strategy. Figures are drawn from the cited third-party sources and reflect information available as of July 2026. This document is intended for institutional investors, qualified purchasers, and financial professionals only.

  2. The content of this blog post is for informational purposes only and is not intended as investment advice, as an offer or solicitation of an offer to buy or sell, or as a recommendation, endorsement, or sponsorship of any security, company, or fund. The information provided does not constitute investment advice, financial advice, trading advice, or any other sort of advice and you should not treat any of the content as such. LYNK Markets does not recommend that any securities should be bought, sold, or held by you. Do your own due diligence and consult your financial advisor before making any investment decisions.