The numbers have become impossible to ignore. Global private credit assets under management crossed $2.5 trillion in the first quarter of 2026, according to data from Preqin and BlackRock’s alternative investments research unit — a doubling from 2022 levels that reflects both institutional demand for yield and a structural retreat by banks from large, complex financing packages.
At the center of this expansion: AI infrastructure.
The Data Center Financing Gap
Building a hyperscale AI data center in 2026 requires between $3 billion and $12 billion in capital, depending on scale, location, and the cost of securing dedicated power. The construction cycle runs 24 to 36 months. The credit profile — long duration, capital-intensive, revenue dependent on hyperscaler lease commitments — fits uneasily into traditional bank loan structures, which have grown more conservative under Basel III Endgame capital rules taking effect in the United States this year.
Private credit fills the gap. Apollo Global Management closed a $5.2 billion data center financing package in February 2026 for a mid-tier colocation operator expanding capacity across four US markets. Blackstone Credit structured a $3.8 billion construction-to-permanent facility for a European AI compute campus in partnership with a sovereign wealth fund. Ares Management has deployed over $8 billion into digital infrastructure credit since mid-2025.
“These deals would have gone to a syndicate of 15 banks five years ago,” said one senior Apollo credit executive speaking on background. “Today, we can close faster, hold more of the ticket, and offer the borrower certainty of execution. The economics work for both sides.”
The Yield Premium — And Its Risks
Private credit loans to AI infrastructure borrowers are currently pricing at SOFR plus 425 to 575 basis points, translating to all-in yields of approximately 10.5% to 12% as of April 2026. Investment-grade corporate bonds in the same sector yield 5.3% to 6.1%. That spread — roughly 500 basis points — reflects illiquidity, structural complexity, and construction risk, but it has proven highly attractive to pension funds and sovereign wealth funds starved for yield in a market where public fixed income offers limited upside.
The California Public Employees’ Retirement System (CalPERS) increased its private credit allocation to 8% of total assets in January 2026, citing AI infrastructure as a key deployment theme. The Canada Pension Plan Investment Board (CPPIB) has committed $12 billion to digital infrastructure debt through 2028.
The risks, however, are accumulating with the capital. Underwriting assumptions for AI data centers depend heavily on hyperscaler lease renewal rates and utilization projections that have not been stress-tested through a full demand cycle. CoreWeave’s March 2026 IPO — which raised $1.5 billion at a valuation below its 2025 private round — revealed that GPU-as-a-service revenue is more volatile than lease-based colocation. Several private credit lenders have begun requiring covenant packages that include minimum utilization floors and dedicated power contracts as conditions of funding.
Banks Are Not Gone, Just Repositioned
Traditional lenders have not exited AI infrastructure entirely. JPMorgan, Bank of America, and Wells Fargo continue to provide revolving credit facilities and investment-grade term loans to the largest hyperscalers — Microsoft, Google, and Amazon — where credit quality and covenant simplicity meet bank requirements. What they have largely ceded is the middle tier: operators too large for small-bank financing and too complex or construction-stage for public bond markets.
That middle tier is precisely where the most aggressive growth in AI compute capacity is occurring. Companies like CoreWeave, Lambda Labs, and Crusoe Energy — building GPU clusters and associated power infrastructure — are structurally dependent on private credit for their capital stacks.
Structural Implications
If AI infrastructure build-out proceeds at the pace implied by current hyperscaler capex guidance — Microsoft alone has committed $80 billion in FY2026 infrastructure spending — private credit’s AI deployment exposure could reach $400 billion by 2028, according to estimates from Morgan Stanley’s alternative lending research team.
At that scale, the private credit market’s health becomes partially correlated with AI adoption rates and data center utilization — a linkage that rating agencies and institutional allocators are only beginning to model. The premium is real. So is the concentration risk.
Discussion
Sign in to join the discussion.
No comments yet. Be the first to share your thoughts.