In plain English, I think this means that if demand for AI drops for some reason, Meta promises to pay Blue Owl some kind of minimum value for that data center.

Hazarding a guess on the accounting without looking at the contract is always speculative. All I have to go by is a relatively spartan press release. With that giant caveat, here are a few guesses on what the reporting might look like and why Meta might want to pay 100 bps more in interest expense relative to what their public debt might cost.

What would the balance sheet of the JV look like?

• The Meta-Blue Owl JV’s assets will be the data center under construction, which will eventually show up at say $27 billion. Equity, which is owned by Meta, would consist of $5.4 billion (20%) stake. Blue Owl will fund part of its capital through debt issued to Pimco and other bond investors via a private securities offering. Morningstar reports, “while S&P Global Ratings assigned the bonds an investment-grade rating of A+, the debt yielded 6.58% at issue – a rate typically associated with high-yield issuances.” Eventually this higher interest cost will of course have to be paid by Meta as there is no free lunch, regardless of how the deal is structured.

How will the JV show up in Meta’s financial statements?

• Meta holds 20% in the JV. The JV is a variable interest entity (VIE). When a company has a relationship with a VIE, it must consolidate that VIE if it is the primary beneficiary. The primary beneficiary is the party that has both (i) the power to direct the activities that most significantly impact the VIE’s economic performance; and (ii) the obligation to absorb losses or the right to receive benefits from the VIE that could be potentially significant.
• This is the gray area. Does Blue Owl (the 80% partner) have the power, or does Meta? Even though Meta is the minority owner, its role as the sole lessee (and the one the data center is being built for) might give it “de facto” power over the JV’s most significant activity—managing that specific asset and lease.
• S&P’s press release on the deal seems to suggest that Meta will not consolidate the full asset and liability on its books. That press release indicates that the JV’s board will be appointed by Blue Owl. So, legally, Meta does not control the JV’s board.
• Where does that leave us? I guess that Meta is likely to follow the equity method of accounting and is hence likely to report only its share of $27 billion when spent or $5.4 billion as a financial investment (not as a data center) in its balance sheet. Crucially, Meta does not have to add the $27 billion in data centers to its PPE (property plant and equipment) or its capex. On top of that, there is a vague reference to Meta contributing land held for sale to the venture. Thus, Meta has substituted its land for an investment entry on its balance sheet.
• Does the residual value guarantee make Meta part owner or a lessee? This deal looks partly like an equity method investment and partly like a lease. If accounted for as a lease, Meta’s guarantee will show up as an off-balance-sheet contingent liability. The sum of the present value of the lease payments to the JV and the expected payout to the JV will be reflected as a right-of-use asset (ROU) on Meta’s balance sheet. The associated lease liability will be added to liabilities if classified as a lease under ASC 842.
• I suspect we will end up seeing both when Meta’s new set of financial statements come out. That is, the equity method investment in the unconsolidated entity (the data center) will show up at 20% of the data center’s costs (around $5.4 billion). The present value of the operating lease payments will show up as a right to use (ROU) asset and an operating lease liability. The residual value guarantee will be reflected as a contingent liability. The ROU will hurt the asset light story a bit but perhaps not to the tune of the full $27 billion had Meta simply borrowed that much to build its own data center without involving Blue Owl. This statement assumes that the present value of operating lease payments plus expected payout Meta makes is smaller than $27 billion.

What does the private credit deal with Blue Owl buy Meta?

• The plain vanilla public debt alternative for Meta was to borrow $27 billion from the public market and build a data center in Louisiana. That transaction would have added $27 billion to capex and its PPE and added $27 billion as debt to its liabilities side. Meta would have paid around 5.5% to borrow that debt. How do we know that? For a comparative transaction, consider the $30 billion of unsecured bonds that Meta raised in the public debt market on October 31, 2025. Fun fact- this was the fifth largest bond issue ever done in US markets!
• An important nuance I owe to Trevor Harris: interest on borrowed debt before the data center is constructed can be added to the cost of the data center. Hence, EBITDA will involve a smaller I (interest) and a slightly higher DA (depreciation and amortization) as the interest capitalized will have to be depreciated eventually.
• The accounting privilege of looking somewhat asset light and liability light is a plus with the private credit deal with Blue Owl. Of course, Meta gets to keep the associated debt of $27 billion off its balance sheet. S&P issued a statement clarifying that they will not consolidate the Meta-Blue Owl JV’s debt with Meta’s debt.
• For this accounting privilege, the Meta-Blue Owl JV is paying 6.58% to borrow money from Blue Owl’s funds. That is at least 100 bps more than what Meta would have paid in the public debt market. The dollar amount is approximately 1% of $27 billion or $270 million per annum.
• To be fair, the Meta press release talks about the speed and efficiency with which private credit operates. The reader can judge whether that feature alone is worth 100 bps a year.

Why potentially pay $270 million a year extra to look asset and liability light?

This is where things become somewhat unclear (at least to me).

• Is it because Meta’s capex spending, year over year, is already quite high? The numbers are $15 billion, $18 billion, $31 billion, $27 billion and $37 billion for the years 2020-2024.
• I could not figure out how much Meta has spent on the Metaverse project and how that spending was split between capex (capitalized as assets on the balance sheet) and opex (operating expenses). In its 10-K, Meta reports segment-based income statements for Reality Labs, the Metaverse business, but does not report a segment-based balance sheet. The losses reported for Reality Labs, year after year, are staggering: $17.7 billion in 2024, $16.2 billion in 2023, $13.7 billion in 2022, $10.1 billion in 2021, $6.6 billion in 2020, and $4.5 billion in 2019. The trail runs cold prior to 2019. These income statement losses cumulate to $69 billion!
• Of course, we cannot figure out the impairment on capex as the Metaverse capex number is not publicly known. If you are Meta, would you really want to entertain more questions about future impairments of AI capex? Perhaps not.
• Off-balance sheet AI investments make such uncertain bets look like opex via lease expenses as opposed to larger capex commitments.
• Meta has minimal on-balance sheet debt of around $29 billion as per their 2024 10-K. Don’t forget the $30 billion bond issue in October 2025. Why would Meta want to load on $25–30 billion of new debt? It’s best to keep that incremental debt off balance sheet, right? To be fair, Meta reported operating cash flow of $91 billion in 2024. That is a lot of money. But tripling on-balance sheet debt in a year ($28 billion to $90 billion) can never be an attractive look to the stock market.

We will never know for sure but perhaps the CFO and the CEO will probably happily pay $270 million a year to avoid questions about capex, potential impairments, leverage and depreciation on Meta’s balance sheet.

Meanwhile, what is Blue Owl up to? We will save that for part 2 of this story.