AI Data Center JV Bond Terms Diverge as Oracle Earns Fitch's Highest Rating
Alina Collins
Five hyperscalers have raised roughly $90 billion in private-placement debt through off-balance-sheet JVs for AI data centers, but Fitch finds investors are substituting brand trust for contractual scrutiny — three Google-backed projects, three different credit outcomes.
$90 billion in JV debt — what are investors actually buying?
Alphabet, Amazon, Meta, Microsoft, and Oracle have issued about $90 billion in private bonds through off-balance-sheet joint ventures to fund AI data centers.
Fitch senior director Anubhav Arora flags the blind spot: investors focus on whether Google or Meta is the guarantor, overlooking that rent payments begin only after construction is complete.
This means → if a data center build is delayed, bondholders face not lower returns but a total cash-flow gap.
Same Google backing — why do the ratings diverge so sharply?
Cloud operator FluidStack's three data-center JV projects issued nearly $11 billion in bonds, all backed by Google lease guarantees — yet terms step down project by project.
Project one: Google fully assumes the lease on default. Project two: the guarantee carries a 25% rent discount. Project three: rent is tied to operating costs, offering the weakest protection.
Construction terms also split — only one of the three has a guaranteed-maximum-price contract; another lets FluidStack and Google exit the lease if delays exceed six months.
In plain terms = if a project stalls, bondholders lose not just rental income but potentially the counterparty itself. Only the Louisiana-based HUT 8 project ($3.25 billion) earned an investment-grade rating.
Oracle is the weakest financially — so why did it score highest?
Among the five hyperscalers, Oracle has the weakest balance sheet; its CDS — credit default swap, a market gauge of default risk — spread has widened sharply in recent weeks.
Yet Oracle's Michigan data-center project includes a "hell-or-high-water" clause: rent is due from a fixed date regardless of whether the facility is built.
This means → Fitch rates contract hardness, not brand size. That single clause eliminates construction-delay cash-flow risk entirely, earning the highest score Fitch has given in this sector.
Can pricing stay one-size-fits-all?
Goldman Sachs credit analysts note that despite wide term differences, most JV bonds trade at nearly identical funding costs, and their prices move in lockstep.
Goldman expects greater yield dispersion as the presence — or absence — of protective covenants begins to matter.
Supply is amplifying the pressure: the five hyperscalers have issued nearly $200 billion in corporate bonds this year; six AI hyperscalers combined have raised about $244 billion, more than double last year's total.
Has the core question for investors changed?
Goldman chief credit strategist Amanda Lynam notes the five hyperscalers have only about $510 billion of additional capacity in the USD investment-grade market — a fraction of the estimated $5.8 trillion in total capex ahead.
This means → JV debt will remain an essential funding tool; the hyperscalers are already tapping euro, Swiss-franc, and sterling markets as well as private credit through firms like Apollo Global Management.
Arora adds that labor shortages and other real-world constraints are raising construction difficulty, potentially giving project owners stronger bargaining power — deal terms will stay bespoke and resist standardization.
In plain terms = the question is no longer "Is AI infrastructure worth investing in?" but which side of the contract you stand on — "who guarantees" and "on what terms" are diverging into two entirely different risk exposures.
Content is for reference only, not financial advice.