Major Divergence in AI Data Center Bond Credit Quality: Oracle Receives Highest Rating from Fitch, While Google-Backed Projects Face Stark Contrasts

The five major hyperscalers have raised approximately $90 billion in private placement bonds for AI data centers through off-balance-sheet joint ventures (JVs). However, the fixed-income market is making a dangerous pricing error: investors are substituting the brand credit of guarantors for a prudent assessment of contractual terms—Fitch Ratings is tearing apart this illusion with differentiated credit scores.

Anubhav Arora, Senior Director at Fitch, told Bloomberg: Investors only look at whether the parent company providing the lease guarantee is Google or Meta, without considering that "rent payments only begin after project completion." If data center construction faces delays, bondholders will face cash flow gaps. Ironically, Oracle, the financially weakest among the five giants, received the highest credit score in this sector to date from Fitch precisely because it committed to "paying rent from the agreed date regardless of whether construction is completed" in one transaction.

Goldman Sachs credit analysts point out that investors tend to view JV project bonds as a homogeneous asset class. However, Citigroup’s analysis of three JV projects for cloud operator FluidStack proves that with the same Google endorsement, credit outcomes can range from investment grade to near junk status—depending on whether guarantee terms include rent discounts, tenant exit rights, and capped construction contracts.

Same Google, Three Credit Outcomes

FluidStack’s three data center JV projects issued a combined nearly $11 billion in bonds, all with lease guarantees provided by Alphabet’s Google, yet the rating results were vastly different.

According to Citigroup analysts: The first project features a comprehensive lease guarantee from Google—if the cloud operator defaults, Google directly assumes the lease; the second project’s Google guarantee comes with a 25% rent discount; and the third project’s rent formula is linked to operating costs. While the latter two guarantees can cover debt service costs, their level of protection is significantly weaker than the first.

More critical differences lie in construction contracts and exit clauses. Only one of the three projects signed a guaranteed maximum price construction contract; another includes a clause allowing FluidStack and Google to exit the lease if construction delays exceed six months—for bondholders, this means that if the project stalls, there would be no rental income, and even the party to pursue could disappear.

Ultimately, only the HUT 8 DC LLC project in St. Francisville, Louisiana (sized at $3.25 billion) received an investment-grade rating. This project benefits from Google’s comprehensive lease guarantee, a capped construction cost contract, and no tenant exit rights.

Oracle’s "Reverse Logic": Weakest Creditworthiness, Highest Score

Among the five major hyperscalers, aside from Oracle, the other four—Alphabet, Amazon, Meta, and Microsoft—all have very healthy cash flows. Yet it was Oracle, with the weakest financial strength and rapidly widening CDS spreads recently, that secured Fitch’s highest credit score in this sector to date, thanks to an unconditional rent payment guarantee for its Michigan data center project.

Arora’s summary points to a core contradiction: Investors look at the guarantor’s brand, while rating agencies look at contractual details. When a project combines "tech giant endorsement" with "tenant exit rights," the former may give investors a fatal false sense of security.

Countdown to the End of Homogeneous Pricing

Although there are significant differences in terms among JV project bonds, financing costs for most projects are highly similar, and bond price movements are closely correlated. Goldman Sachs credit analysts expect yields to diverge further as the importance of protective clauses—or the lack thereof—becomes increasingly prominent.

Pressure for divergence is being amplified by the supply side. The five giants have issued nearly $200 billion in corporate bonds this year, and the six AI hyperscalers have cumulatively issued about $244 billion in bonds year-to-date, more than double last year’s total. Amanda Lynam, Chief Credit Strategist at Goldman Sachs, noted that the five giants have only about $510 billion in additional issuance capacity in the USD investment-grade bond market—a tiny fraction relative to Goldman Sachs’ estimated $5.8 trillion in total capital expenditures.

This means JV bonds will remain an indispensable financing tool. Tech giants have already issued bonds in multi-currency markets such as the euro, Swiss franc, and pound sterling, and leveraged private credit channels from institutions like Apollo Global Management. Meanwhile, demand for data centers has become so strong that developers are beginning to face real-world constraints such as labor shortages. Arora pointed out that this may grant landlords greater bargaining power, implying that transaction terms will remain highly personalized rather than converging toward standardization.

For fixed-income investors, the core question is no longer "Is AI infrastructure worth investing in?" but rather "Which side of the terms are you on?"—Who provides the endorsement, and under what conditions, is diverging into two distinctly different risk exposures.