Edward Obuz on Private Credit and the AI Buildout

Edward Obuz just read the Social Capital deep dive “Is the AI Buildout Creating a Private Credit Bubble?” Here is the commentary.

The AI infrastructure race is no longer funded mainly by operating cash flow and equity. Debt, especially private credit, is carrying a growing share of the load. That shift is real, the numbers are large, and the structures deserve scrutiny. Whether it amounts to a bubble is a separate question that depends on how the debt is underwritten, how the mismatches are managed, and whether the underlying AI cash flows materialize on the timelines lenders assume.

Edward Obuz on How Private Credit Is Funding the AI Buildout

Private credit is direct lending between non-bank lenders and borrowers. It expanded after 2008 as banks pulled back from parts of corporate lending. BIS data show U.S. direct loans growing from roughly $96 billion in 2010 to about $2.4–2.5 trillion by 2025. Lending to software and technology companies grew even faster, from about $22 billion to more than $1 trillion over the same period. Those figures check out against the Bank for International Settlements publications.

The AI buildout has accelerated the trend. In the second quarter of 2026, Apollo and Blackstone led a roughly $35 billion vehicle to purchase chips (primarily Google TPUs developed with Broadcom) and lease them to Anthropic. UBS and multiple outlets described it as one of the largest private credit deals on record. For context, PitchBook data cited in the Social Capital piece put traditional direct lending to private-equity-backed U.S. companies at about $23 billion in the same quarter. The structure keeps the hardware in a special-purpose vehicle, off Anthropic’s balance sheet, with Broadcom providing residual-value support on senior tranches. That is standard structured finance adapted to compute.

S&P Global Ratings has projected that Alphabet, Amazon, Meta, Microsoft, Oracle, and SpaceX (xAI) will spend about $870 billion in 2026 and more than $1.3 trillion in 2027 on AI infrastructure, with debt, leases, and other financing filling the gap once free cash flow turns negative. Other forecasts from Goldman Sachs and Morgan Stanley sit in a similar range once SpaceX is included. The absolute numbers are large relative to prior technology investment cycles.

What Edward Obuz Sees in the Pricing and Structure Risks

BIS analysis found that private loans to AI-related companies carried an average spread of 6.2 percentage points above the base rate in early 2026, versus 6.1 points for other private credit loans. The near-identical pricing suggests lenders are not charging a material premium for AI-specific risks, or that equity markets are pricing more aggressive growth. Both can be true. Private credit is also exposed to the software companies AI may disrupt. If generative AI reduces enterprise software revenue faster than expected, existing software borrowers come under pressure. If AI revenue ramps more slowly, the new AI infrastructure borrowers face shortfalls when loans mature.

Duration and contract mismatches add another layer. CoreWeave has borrowed for five years against customer contracts averaging closer to three years. Meta’s Hyperion bonds run to 2049 against leases that renew every four years. These are not hidden; they are features of project finance. They work if the underlying demand and residual values hold. They amplify losses if they do not.

Secondary-market pricing already differentiates. Debt financing chips leased to xAI traded up after SpaceX’s acquisition. Project Jupiter, the roughly $18 billion financing for an Oracle-leased data-center campus in New Mexico, later traded at 89–91 cents on the dollar amid local opposition, permitting delays, a force-majeure notice, and broader concerns about Oracle’s leverage. Those marks are consistent with Financial Times and Reuters reporting.

Edward Obuz View on Whether This Is a Private Credit Bubble

In my view, private credit is not a monolithic bubble. The post-2008 growth was driven by a genuine gap in middle-market and technology lending, and the asset class has absorbed capital while maintaining relatively low leverage at the fund level. The newer AI infrastructure slice is different. It is larger, more concentrated in a handful of counterparties and projects, and more dependent on long-term assumptions about AI monetization, power availability, and residual hardware values. Spread compression and limited differentiation between AI and non-AI private loans are the clearest yellow flags. Net outflows from some private credit funds in early 2026, noted in the Social Capital piece, are another early signal of investor caution.

The practical distinction is underwriting quality and structural protection, not the label “private credit.” Deals with strong residual-value support, investment-grade senior tranches, and diversified offtake look more like infrastructure project finance. Deals that rely on aggressive growth assumptions, thin equity, and short-term contracts rolled into long-term debt look more like the leveraged software loans that are already showing stress. Lenders and limited partners who treat the two as interchangeable will discover the difference in the next refinancing cycle.

The organizations that will navigate this well are the ones that underwrite the specific cash-flow and residual risks rather than the AI narrative. That is a credit skill, not a technology skill. For related analysis on how AI is already reshaping private credit risk management, see Navigating the 2026 Private Credit Crisis.

Edward Obuz is a Toronto-based commentator focused on AI strategy and capital markets. He reviews primary research and market data on technology financing. His work centers on how AI infrastructure spending is funded and the resulting credit risks. That experience fits an analysis of private-credit structures in the current buildout. The practical problem the perspective addresses is distinguishing underwritten project risk from narrative-driven leverage in AI-related debt.

Further reading: Edward Obuz on the financial architecture of AI giga-factories and the full author profile on edwardobuz.com.