Newsletter · 2026-09-02 · Ashutosh Agarwal

The Off Balance Sheet Vehicle Behind Meta's Data Center Turns Out to Be Blue Owl - Private Credit & Alternatives - Week of September 2, 2026

Private Credit & Alternatives for the week of September 2, 2026. Podcast synthesis on the Blue Owl owned joint venture financing Meta's Hyperion data center with $27 billion of bonds, Bloomberg Intelligence explaining the SPV chip-leasing structure behind Broadcom's $60 billion and NVIDIA's $500 billion deals, and a Texas apartment operator describing how equity gets wiped while debt stays liquid.

Private Credit & Alternatives

Week of September 2, 2026: The Off Balance Sheet Vehicle Behind Meta's Data Center Turns Out to Be Blue Owl

This week closes a loop that has been open for weeks. For most of the summer, the bull case and the bear case on the AI-lending boom agreed on one strange, specific point: the borrowed money is not sitting on the balance sheets of the cash-rich technology giants. It is downstream, in what everyone kept calling, vaguely, "special purpose vehicles," off-the-books financing structures built to fund the data-center build-out. The bull (Raoul Pal, two weeks ago) said those vehicles were where any blow-up would happen. The bears said the same, and added: that is exactly where the private-credit and insurance money went. But nobody had named one.

Two mid-August conversations that had not yet surfaced in this newsletter finally did the naming. One walks through, layer by layer, the actual structure financing one of Meta's largest data centers, and the majority owner turns out to be Blue Owl, one of the marquee private-credit firms whose own funds are now reporting five-year-high defaults. The other, from a credible Wall Street analyst rather than a pundit, explains the plumbing of Broadcom's new $60 billion AI debt deal and NVIDIA's $500 billion one: who buys the chips, who leases them, and why the whole thing is designed to keep the borrowing off the balance sheets of the companies you would worry about. The abstraction from the last month now has concrete names, dollar figures, and a diagram. That is the story of the week.

TL;DR

The "special purpose vehicle" everyone kept warning about has a name, and it is a private-credit giant. On The Banker Next Door, the host walked through a Wall Street Journal analysis of the structure behind Meta's "Hyperion" data center: a joint venture owned "20% Meta, 80% Blue Owl," with "$27 billion in debt" raised in a bond sale, none of it on Meta's balance sheet. Across the big tech names, the Journal counted "$1.2 trillion in off-balance-sheet obligations" from leases that have not started yet, "about four times more than what was disclosed a year earlier."
A Wall Street analyst explains why NVIDIA can bankroll the whole thing and never run out of cash. On Bloomberg Intelligence, Bloomberg's global head of tech research, Mandeep Singh, described Broadcom's "$60 billion of debt" deal and NVIDIA's "$500 billion" package with "six private equity players" as the same trick: an SPV buys the chips and leases them to a lab like Anthropic or OpenAI, so the borrower "doesn't have to carry that on its balance sheet." NVIDIA, generating "almost $100 billion plus in free cash flow every year" at "75% gross margins," can afford to backstop all of it.
Down in real estate, the credit market is sending a preview: debt is easy, equity is being wiped out. On the Old Capital Real Estate podcast, veteran Texas apartment operator Michael Becker said "the debt market is extremely robust and liquid" while "equity is wiped in a lot of these cases," with lenders now "taking a discounted payoff or… foreclosing" on workforce-housing deals, a live rehearsal of what happens to the riskiest slice when a credit cycle turns.

What's new

1. The off-balance-sheet vehicle finally gets a name, and it is Blue Owl. For weeks, both sides of the AI-debt argument have pointed to the same shadowy place, the "special purpose vehicles" financing the data-center boom, without anyone laying one open. On The Banker Next Door (Aug 20), the host did exactly that, reading through a Wall Street Journal breakdown of the structure behind Meta's Hyperion data center. Treat the show's editorial temperature accordingly, this is a dramatic, retail-facing channel, and by the end the host is comparing the tech giants to Thelma and Louise "driving themselves off the cliff." But the numbers underneath come from the Journal's reporting on the companies' own disclosures, and they are the useful part.

Here is the structure, in plain terms. Meta wanted a giant data center (Hyperion) but did not want the debt for it on its books. So a joint venture was created, owned "20% Meta, 80% Blue Owl," and "though Meta is the builder, neither Hyperion nor the $27 billion in debt that's financing its construction show up on Meta's balance sheet." A Blue Owl holding company that owns the stake "raised the construction financing in a bond sale," so the ultimate lenders are bondholders. Then the money circles: "Meta basically sends payments to the special venture and then they send the payments out to the bondholders." Meta is, in the host's phrase, "Hyperion's minority partner and tenant" at once, its lease payments are the cash flow that pays the bonds.

Two details make this more than a curiosity. First, the accounting is deliberately invisible: Meta "initially agreed to lease Hyperion for a four-year term starting in 2029 with options to renew for up to 20 years," and "guaranteed that it would make bondholders whole if it doesn't stay the entire two decades," yet because it does not judge those payments "probable," it "hasn't recorded any liability on its balance sheet." Its stated "aggregate initial lease commitment is about $12.3 billion," and it disclosed "$347 billion in total obligations for leases that haven't kicked in yet." Second, this is not a one-off. Across the companies the Journal reviewed, Alphabet, Amazon, Microsoft, Oracle, NVIDIA, Broadcom, SpaceX and AMD, plus Meta, "promises of payments under these uncommenced leases total $1.2 trillion in off-balance-sheet obligations, or about four times more than what was disclosed a year earlier." Add long-term purchase commitments to lock in chips and hardware and the figure balloons: "those and other purchase obligations… stand at a whopping $1.9 trillion." Alphabet's alone "stood at $811 billion as of June 30th," up from "$332 billion… three months earlier," with some energy agreements running "as far out as 2054." NVIDIA, separately, "committed to make $27 billion in equity investments" by the end of its fiscal year in January 2027.

Why this matters for us specifically: the private-credit angle is no longer theoretical. Blue Owl, a name that has appeared in this newsletter as one of the funds whose own quarterly reports show defaults at five-year highs, is here on the other side of the ledger, as the majority owner of the vehicle bankrolling Meta's build-out and the party that went to the bond market to raise the money. This is precisely the "the debt lives downstream, in SPVs" claim made concrete: the downstream vehicle is a private-credit-sponsored JV, and the way it raised its money was to sell bonds to whoever would buy them. The host's alarmism aside, the sober version of the concern is real and was voiced by a Morgan Stanley accounting analyst quoted in the piece: "It is becoming increasingly difficult for investors to assess companies' total potential leverage." When a trillion dollars of obligations sits in structures designed not to appear on anyone's balance sheet, that is not a rhetorical worry. It is an accounting fact.

2. How the chip-leasing machine actually works, from an analyst, not a pundit. The single most clarifying voice of the fortnight belongs to someone who does this for a living. On Bloomberg Intelligence (Aug 21), Bloomberg's global head of technology research, Mandeep Singh, calmly explained the mechanics that the more excitable shows only gesture at, and it is worth separating his measured, insider read from the pundit chorus.

Start with the new deal in the headline: Broadcom "specifically the $60 billion of debt." What makes it notable, Singh said, is that it involves "doing debt of different parts of the capital stack," not one simple bond, but layered financing. And the structure is the same one NVIDIA used with its "around $500 billion" package and "six private equity players." Here is the trick, in his words: "the special purpose vehicle structure… will buy the chips from Broadcom in this case and then lease it to Anthropic." Why lease rather than sell? Because "Anthropic doesn't want to pay for all of that up front so that it doesn't have to carry that on its balance sheet." Anthropic, he noted, "is a fast-growing lab… at a very high trajectory, but doesn't have the balance sheet of the hyperscalers." The SPV absorbs the capital cost; the lab pays rent. NVIDIA plays the same game at the other end: "NVIDIA is using its balance sheet along with the private equity players to set up this special purpose vehicle, which will buy the chips and then lease it out to… users like OpenAI or other companies."

The part that should reassure the bulls, and it is the strongest single argument for why this does not detonate at the top, is Singh's answer to whether NVIDIA itself could ever go cash-flow negative building all this, the way Alphabet just did ("Alphabet… went negative free cash flow for the first time this quarter," he confirmed). His answer was flat: no. NVIDIA is "generating… almost $100 billion plus in free cash flow every year," it has "become what Apple has done over the past decade," and "with 75% gross margins, they will continue to do that in the foreseeable future." Crucially, "they're not setting up their own fabs. So the only investment they have is really in terms of backstopping a lot of these SPV structures." A company throwing off $100 billion a year, with no factories to fund, can comfortably act as the shock absorber for the whole financing chain.

Hold Singh's two points together and you have the cleanest statement yet of the cycle's actual risk shape. The chip vendors (NVIDIA, Broadcom) are financially bulletproof and are deliberately using that strength to keep the debt off the balance sheets of the borrowers who are not: the labs, via SPVs, funded by private equity and the bond market. That is not a bug the industry is trying to hide; it is the design. The reassuring reading is that the strongest players are backstopping the weakest, so nothing breaks. The worried reading is the mirror image: the strength at the top exists precisely because the fragility has been pushed, by design, into off-balance-sheet vehicles owned by private-credit firms and sold to bondholders. Both readings are true descriptions of the same structure. Singh, notably, offered no opinion on which one wins, he just described the machine. That neutrality, from the most credible voice of the fortnight, is its own kind of signal.

3. A ground-level preview from real estate: debt is easy, equity is getting wiped. While the AI-credit story plays out in trillions and SPVs, the older, humbler corner of private lending, commercial real estate, is quietly running the experiment everyone in private credit should be watching: what actually happens to the different layers of a deal when a credit cycle turns. Two operators described it this week, and both are insiders who lend and buy for a living, not commentators.

On the Old Capital Real Estate podcast (Aug 20), Michael Becker, who owns and operates roughly 20 apartment complexes across Dallas–Fort Worth, drew the sharpest possible line between the two halves of a capital structure. "The equity market is tough, but the debt market is extremely robust and liquid," he said: Fannie Mae, Freddie Mac, bridge lenders, life-insurance companies and banks are all "throwing money" and "these spreads are historically pretty tight… you can get debt on almost anything." Equity is the opposite. Investor sentiment, he said, is "about as bad as I've felt since probably 2009 or 2010"; the "workforce housing syndicated deals from the peak have lost money or [investors are] afraid they're about to lose money." And the resolution is brutal and specific: "equity is wiped in a lot of these cases, and the lender is now taking a discounted payoff, or they have already foreclosed and they're going to sell at a loss to their debt basis." Lenders, he said, "are forcing people to meet the market in real time." His own book is turning up, DFW "absorbed almost 20,000 units" in the first half of 2026, the most in the country, and his portfolio occupancy is back above 95%, so he thinks the bottom is "now-ish." But the mechanism he describes is the one that matters for private credit generally: when values fall, the equity is destroyed first and the lender either takes a haircut or takes the keys. That is exactly the sequence a private-credit fund's own equity tranche faces if its borrowers sour.

The complementary note came from The Mobile Home Park Exchange (Aug 19), where Art Tuverson, a mortgage banker at Berkadia, explained how one asset class dodged the trouble Becker describes. Manufactured-housing deals are now "being acquired at negative leverage," the all-in cost of the loan is higher than the yield on the property, which normally spells danger. But the sector "really dodged the maturity wall," he said, because the floating-rate loans taken out in 2021–22 were rescued in 2024–25: roughly "25%… were able to refi… with a fixed-rate loan through the agencies" and "about 75%… through CMBS," which was offering five-year, full-term interest-only loans. The takeaway across both: the debt markets have stayed open and generous, which is what has kept a wave of forced selling from becoming a flood. The risk is not that lenders slam the window shut, it is that, as Becker's workforce-housing example shows, when the window stays open at the debt level, it is the equity underneath that quietly gets erased.

The debate

Does pushing AI debt into private-credit SPVs reduce the risk, or just hide it? This is the week's real argument, and for once we can debate the actual structure rather than a rumor of it. The reassuring case is strong and was made implicitly by Mandeep Singh: the chip vendors backstopping these vehicles are the most cash-rich companies on earth, the labs get to grow without drowning in capital costs, and there are "legitimate lenders" underwriting each deal. Financing a customer to buy your product is ordinary commerce. The worried case does not dispute any of that, it disputes the visibility. When Meta can commit to a 20-year lease, guarantee bondholders, and still record no liability because it deems the payments "not probable," and when a trillion dollars of such promises sit off the balance sheets across the sector, the honest conclusion is Morgan Stanley's: outsiders can no longer measure the true leverage in the system. Risk that cannot be measured is not the same as risk that has been removed. The synthesis: the structure genuinely does place the debt with parties who can, today, service it. What it also does is make the total impossible to see, and the last time an entire market ran on obligations nobody could add up, the problem was not that the risk did not exist. It was that no one knew where it was until it moved.

Is the real-estate playbook a preview of private credit's, or a different animal? Becker's description, debt liquid, equity wiped, lenders taking discounted payoffs, is exactly the sequence a stressed private-credit book would run, just in an asset class with decades of workout history and deep agency and CMBS backstops behind it. The bull reading is that this is precisely why it is not a preview: commercial real estate has Fannie, Freddie and a mature distressed-debt ecosystem to absorb the losses in an orderly way, and the fact that even beaten-up workforce housing is clearing through discounted payoffs rather than fire sales shows the machinery works. The bear reading is that the direct-lending world has none of that plumbing: no government-sponsored backstop, no liquid secondary market for a defaulted middle-market loan, and equity tranches inside funds sold to retail buyers who cannot be foreclosed on cleanly. If "equity gets wiped first, lender takes a haircut" is the universal rule, the question is only whether private credit's version resolves as orderly as real estate's, or messier because the shock-absorbers are not there. This week gave us the mechanism; it did not tell us which ending applies.

Who is still not answering. The insurers, Athene, Corebridge, Global Atlantic, F&G and Brookfield Reinsurance, have now gone a fifth straight week without addressing the charge, aired repeatedly this summer, that pension and insurance buyers "don't even know what they bought." The operators running actual direct-lending money have not engaged either. After a stretch in which the first real default numbers landed and the bears finally got their data, the pattern from the last month holds: the loudest voices are the pundits and the sellers, while the balance-sheet holders keep their heads down.

The names in play

Blue Owl (OWL): the week's key name. Per the Wall Street Journal analysis aired on The Banker Next Door, Blue Owl owns 80% of the joint venture behind Meta's Hyperion data center, and a Blue Owl holding company raised the "$27 billion" of construction debt through a bond sale, putting the same firm whose funds are reporting five-year-high defaults squarely on the financing side of the AI build-out.
Meta (META): the tenant-and-minority-partner in Hyperion, with a stated "$12.3 billion" initial lease commitment, a 20-year potential term, a make-whole guarantee to bondholders, and "$347 billion" in leases-not-yet-started kept off its balance sheet.
NVIDIA (NVDA) and Broadcom (AVGO): the chip vendors financing their own demand. NVIDIA's "$500 billion" package with six private-equity firms and Broadcom's "$60 billion" layered debt deal use the same SPV-buys-and-leases structure. Per Mandeep Singh, NVIDIA's "$100 billion plus" annual free cash flow and "75% gross margins" let it backstop the vehicles without ever going cash-flow negative.
Anthropic and OpenAI (private): the labs on the receiving end, leasing chips through SPVs precisely so the capital cost never touches their balance sheets, because, per Singh, they "don't have the balance sheet of the hyperscalers."
The six private-equity backers of NVIDIA's package (Apollo, BlackRock, Brookfield, Goldman Sachs, KKR and others): the alternative-lending giants supplying the SPV capital, the same roster underwriting the frontier's debt from the previous NVIDIA consortium.
Alphabet (GOOGL) and Amazon (AMZN): cited as having gone free-cash-flow negative, Alphabet "for the first time this quarter," as capital spending outruns operating cash. Alphabet's off-balance-sheet purchase commitments reached "$811 billion."
The agencies and CMBS market (Fannie Mae, Freddie Mac): the plumbing that kept commercial real estate from a forced-selling wave, refinancing the bulk of stressed floating-rate loans, the backstop private credit conspicuously lacks.
The insurers, Athene, Corebridge, Global Atlantic, F&G, Brookfield Reinsurance (various): silent for a fifth straight week, still holding the most opaque private-credit risk and still not talking.

Read-throughs

The "SPV" abstraction is now a named, ownable risk, so go look at your Blue Owl exposure through both lenses. The same firm can be a source of default headlines (its direct-lending funds) and a source of AI-financing exposure (the Meta/Hyperion JV). If you own the alternative managers, you now know the AI build-out sits inside them, not just around them. Map which of your holdings are sponsoring these off-balance-sheet vehicles, because that is where the leverage the Morgan Stanley analyst cannot measure actually lives.
Watch the accounting, not just the defaults. The single most important fact this fortnight is that "$1.2 trillion" of lease obligations and "$1.9 trillion" of purchase commitments are structured to stay off balance sheets. The tell will not be a missed payment first, it will be a change in how these obligations get disclosed, or a rating agency forcing them onto the books. Track disclosure and accounting-rule commentary as a leading indicator, ahead of the default data.
NVIDIA and Broadcom's cash strength is the load-bearing assumption, and it is genuinely sturdy. Singh's point that NVIDIA cannot go cash-flow negative is the strongest single reason this does not break at the top. The risk was never the chip vendors; it is the leased-to borrowers and the bondholders funding the SPVs. Keep stress-testing the labs and the vehicles, not NVIDIA's balance sheet.
Real estate is the live experiment for how a credit turn resolves. Becker's "debt liquid, equity wiped, lenders take the haircut" is the sequence to expect if private-credit borrowers sour, but real estate has agency and CMBS backstops that direct lending does not. The read-through: an orderly workout in commercial real estate is the best case for private credit, not the base case. If even that market, with all its shock-absorbers, is wiping out equity and forcing discounted payoffs, the same dynamic in an asset class without those backstops is unlikely to be gentler.
Treat the insurers' continued quiet as information. Five weeks without a response from the firms holding the most opaque risk is a pattern, not a scheduling accident. Until the people actually running direct-lending and insurance money start answering the questions the data raised this summer, assume the absence of reassurance is not the same as reassurance.

What changed

Last week the story was numbers: the first hard default data arrived, five-year highs at the biggest funds, and the bulls and bears agreed the AI debt was hiding downstream in "special purpose vehicles" without anyone naming one.

This week the story is names. The vehicle behind Meta's Hyperion data center turns out to be 80% owned by Blue Owl, financed with $27 billion of bonds, and kept entirely off Meta's balance sheet, one concrete instance of the abstraction everyone had been circling. A Bloomberg analyst then explained the general machine: chip vendors and private-equity firms set up SPVs that buy the chips and lease them to cash-poor labs, so the borrowing never lands where it would scare anyone, while NVIDIA's $100-billion-a-year cash flow quietly backstops the lot. And from real estate came a preview of how a credit turn actually resolves, debt stays liquid, equity gets wiped, lenders take the haircut, a sequence private credit has not yet had to run at scale.

What did not change is who is talking. The insurers stayed silent for a fifth week, the direct-lending operators stayed out of the conversation too, and the loudest voices remained the pundits and the sellers. The summer's central question, are the values real, and where exactly does the risk sit, got a partial, valuable answer this fortnight: we now know the risk sits in off-balance-sheet vehicles owned by private-credit firms and funded by bondholders, and we know a trillion-plus dollars of it is structured so no outsider can measure the total. Knowing where the risk lives is progress. It is not the same as knowing it is safe.