Private Credit & AI Infrastructure

Private credit and AI infrastructure refers to the financing layer underneath the AI-data-center buildout, in which the marginal dollar of data-center construction in 2025–26 increasingly came from private credit — direct lending by non-bank asset managers — channelled through special-purpose vehicles (SPVs) and joint ventures that keep the associated debt off the operating company's own balance sheet. It is distinguished from hyperscaler self-funding, in which Big Tech paid for data centers out of operating cash flow, and is one strand of the AI Bubble Debate. It addresses the financing question that AI Infrastructure Capex hands downstream.

Snapshot

Background

Three pressures pushed the buildout off-balance-sheet and toward private credit.

The first is scale beyond cash flow. Aggregate frontier-lab and hyperscaler capex is running at $700B–$1T in 2026, and Apollo projects roughly $3T cumulative AI-infrastructure capex across the US economy through 2028 (How the AI revolution has turbocharged M&A). Microsoft, Alphabet, Amazon, and Meta, despite large operating cash flow, cannot fund this pace internally without straining their balance sheets and equity stories.

The second is balance-sheet management. Structuring data centers as SPVs or joint ventures lets the sponsor keep the construction debt off its own consolidated balance sheet, preserving credit ratings and reported leverage. Meta's roughly $29–30B Hyperion data-center vehicle in Louisiana, structured with Blue Owl Capital and PIMCO in 2025, is an example: most of the debt sits in the SPV, not on Meta's books.

Stijn Van Nieuwerburgh's Brookings Papers on Economic Activity paper (Fall 2026) uses Hyperion as its worked case. Meta initially owned the roughly 2.0 GW, $30 billion project outright, then sold an 80% equity stake to Blue Owl for approximately $2.5 billion; the joint venture, Beignet, raised $27 billion of debt in October 2025, rated A+ by Standard & Poor's, one notch below Meta, which the paper describes as the largest individual investment-grade corporate debt issuance in U.S. history. The project's debt-to-asset ratio is about 90% and its debt service coverage ratio 1.12×, against Meta's own year-end 2025 leverage of about 25% on book equity and 4% on market equity. Meta holds five four-year leases running from 2029 to the bonds' 2049 maturity, may terminate at each renewal, and must then cover any shortfall between the sale proceeds and a contractual minimum value; under current GAAP neither the future leases nor that residual value guarantee appear on Meta's balance sheet before they take effect. The debt priced at 6.58%, at least 100 basis points above comparable Meta unsecured debt, which the paper puts at more than $5 billion of additional interest over the financing's life, and it describes Meta's rationale as "not cheaper financing, but preservation of an asset-light corporate profile" (Financing the AI Buildout (Van Nieuwerburgh, BPEA Fall 2026)). The paper also states that the five largest hyperscalers' combined capital expenditure is projected to exceed their combined operating cash flow for the first time in 2026, which it presents as the reason financing is moving off their balance sheets.

The third is yield-hungry capital. Private-credit funds raised large pools through the low-default years and need to deploy them; investment-grade-adjacent, asset-backed AI-infrastructure lending against long-dated hyperscaler lease and power-purchase-agreement commitments is one destination for that capital.

Principal lenders

Blue Owl Capital is among the most active AI-infrastructure private lenders. It anchored Meta's Hyperion SPV and made a roughly $3B investment in OpenAI's SoftBank Group/Oracle Stargate project in November 2025 (Source: theinformation.com). It manages around $300B or more in assets.

KKR, Apollo, Blackstone, PIMCO, and Brookfield are large alternative-asset managers building AI-data-center and digital-infrastructure debt books. Apollo's projection of $3T cumulative capex through 2028 is one of the most-cited sizing figures for the buildout (How the AI revolution has turbocharged M&A).

A KKR-led group launched Helix Digital Infrastructure on June 11, 2026 with more than $10 billion in committed capital to finance AI data-center build-out; the Kuwait Investment Authority, Nvidia, and utility Vistra are anchor investors, former AWS CEO Adam Selipsky leads the company, and KKR's Waldemar Szlezak serves as CIO. The company projects the AI data-center buildout requires roughly $15 trillion in additional capital — a sizing figure on the order of Apollo's $3T-through-2028 projection extended across a longer horizon (Source: reuters.com; theinformation.com). The structure — an Nvidia-anchored, sovereign-backed financing vehicle — echoes the vendor-guaranteed-lease arrangements Nvidia was reported to be extending to OpenAI's Ohio site the same week (see Data Center Siting / AI Power Politics).

BlackRock acts both as a lender and as an equity consolidator; its roughly $39B AES/Allete utility transaction connects the private-credit thread to the power-supply thread (How the AI revolution has turbocharged M&A).

Hyperscaler equity counterpoint

Not all financing is private credit. On June 1, 2026, Alphabet said it plans to raise up to $80B in equity to fund its AI buildout: $30B in underwritten public offerings (mandatory convertible preferred and common), a $40B at-the-market program starting in the third quarter, and a $10B private investment from Berkshire Hathaway, extending a push that already made Alphabet the first modern company to issue a 100-year bond (Source: axios.com). Sundar Pichai framed the demand-side logic for the spending by stating that "the risk of under-investing is dramatically greater than the risk of over-investing" (Source: axios.com). Alphabet's funding through public equity and a strategic investor contrasts with Meta's and OpenAI's funding through private-credit SPVs, a difference attributed to balance-sheet strength and risk appetite.

Debates and positions

Commentators describe private-credit financing as changing the shape of AI-buildout risk, not only its size, along three lines.

On opacity, SPV and private-credit structures are less transparent than public-company balance sheets, so the true leverage of the buildout is hard to observe. Van Nieuwerburgh argues that the aggregate equity-debt mix understates leverage because hyperscalers put internal capital into IT equipment while shells, power and real estate are financed with project debt and leases, raising loan-to-value on the infrastructure even as reported hyperscaler leverage stays low. He notes that publicly rated deals such as Hyperion are a relatively small share of data-center financing, with much larger volumes in syndicated bank loans and private-credit facilities "for which little information is publicly available," and proposes disclosure mapping project-level debt, joint-venture interests, future lease commitments, residual-value and construction guarantees, tenant concentration, renewal options and assumed recovery values. He writes that it "would be premature to conclude that AI infrastructure already poses systemic risk comparable to earlier credit booms," but that off-balance-sheet structures "may make correlated exposures harder to observe before a downturn" (Financing the AI Buildout (Van Nieuwerburgh, BPEA Fall 2026)).

On liquidity, private-credit funds promising periodic liquidity have shown stress: Blue Owl drew scrutiny in early 2026 after gating redemptions on a fund, which rattled the broader roughly $1.8T private-credit market (Source: advisorhub.com). If AI revenues disappoint while construction debt matures, the mismatch between long-dated illiquid AI-infrastructure loans and investor redemption expectations is a candidate transmission channel for stress.

On circularity, several deals route capital among the same set of actors, with chipmakers investing in labs that lease capacity from clouds financed by lenders backstopped by chip-demand assumptions, a pattern the AI Bubble Debate identifies as a warning sign.

Ed Zitron argued in a July 22, 2026 essay that AI data-center financing is accumulating subprime-style systemic risk, comparing layered debt claims on the same underlying assets to the synthetic CDOs of the 2008 crisis (Source: wheresyoured.at).

The private-credit thesis is contested. Bulls argue the financing is rational asset-backed lending against contracted hyperscaler demand; bears argue it is the kind of opaque, leveraged, circularly-financed buildout that precedes infrastructure overbuilds. Both readings depend on whether AI inference revenue materializes at the pace the capex assumes (AI Infrastructure Capex; AI Bubble Debate).

The June 2026 Broadcom–Apollo–Blackstone "Big Sky" fund is a representative structure: an initial ~$35 billion commitment led by Apollo to finance more than 20 gigawatts of Broadcom-designed capacity through 2028, split across three debt tranches — $6 billion to banks, $24 billion to asset-backed investors at a 5.75% yield, and $4.4 billion of junior debt at 8.5% — with the two senior tranches backstopped by Broadcom and the transaction arranged by Morgan Stanley. The fund finances Anthropic's lease of Alphabet-designed TPUs that Broadcom manufactures, alongside OpenAI-tied projects, layering chip-vendor credit support onto the asset-backed structure (Source: ft.com).

A parallel development is the early financialization of compute itself: Goldman Sachs and JPMorgan were reported in June 2026 to be exploring futures contracts tied to GPU rental prices, with Intercontinental Exchange, CME Group, DRW, and the CFTC among the parties involved and exchange-listed products expected later in 2026 — a step toward treating compute as a tradable, hedgeable commodity rather than a purely contracted input (Source: theinformation.com). CME Group and Silicon Data subsequently set October 5, 2026 for the launch of two cash-settled GPU rental-index futures, and the CFTC opened a request for comment on August 19, 2026 on whether compute can support such a contract at all; see Compute derivatives.

The financing channel widened further in mid-2026. Morgan Stanley, one of the most active banks financing data-center developers, began pitching some of those clients on June 21, 2026 to raise money in the leveraged-loan market — a channel better known for funding leveraged buyouts — as buildout costs pressed developers toward new capital sources; collateralized-loan-obligation managers were reported to be showing strong appetite for data-center leveraged loans (Source: theinformation.com). Routing data-center debt through the leveraged-loan and CLO markets extends the buildout's financing beyond direct private credit into syndicated, tradable instruments, broadening the set of investors exposed to AI-infrastructure credit risk.

Relationships

• depends-on: AI Infrastructure Capex
• related: AI Bubble Debate
• related: AI Data Centers
• related: Constellation Energy
• supports: How the AI revolution has turbocharged M&A
• related: Financing the AI Buildout (Van Nieuwerburgh, BPEA Fall 2026) — Hyperion (Beignet) worked example; asset-level leverage and disclosure argument
• related: Stijn Van Nieuwerburgh