Meta's $27 Billion Hyperion Bond: Who Owns the Asset and Who Carries the Risk

Meta's Hyperion campus in Louisiana is funded by $27.3 billion of bonds that Meta did not issue and does not report as debt. The issuer is a Blue Owl vehicle called Beignet Investor, and PIMCO anchored the 144A sale. Meta data center financing on this pattern moves the asset off the balance sheet, but Meta's own filing still puts its maximum exposure at $46.03 billion.

Meta data center financing for Hyperion, in one page

The structure has four parties and one asset. Meta and funds managed by Blue Owl Capital signed a joint venture agreement on 21 October 2025 to develop and own the Hyperion campus in Richland Parish, Louisiana. Blue Owl's funds hold 80% of that venture and Meta holds 20%.

The two partners committed to fund their pro rata share of approximately $27 billion in total development costs, according to Meta's announcement. Meta contributed land and construction work already under way. Blue Owl's funds put in about $7 billion of cash, and Meta took a one-time distribution of about $3 billion back out.

The debt does not sit in the joint venture itself. A separate Blue Owl subsidiary, Beignet Investor, sold $27.3 billion of 6.581% senior secured notes due May 2049 in a 144A and Reg S format. Morgan Stanley was sole underwriter and PIMCO was the main anchor investor, according to IFR.

If the $29 billion figure sounds more familiar, it is the August 2025 headline. Early reports described about $26 billion of debt led by PIMCO and $3 billion of equity from Blue Owl. The final bond came in at $27.3 billion, so treat the $29 billion as the negotiating number rather than the closing one.

Who owns the Hyperion asset, on paper and in practice

On paper, the answer is Blue Owl. Its funds hold 80% of the equity and, according to IFR's breakdown of the deal, the formal governance of the joint venture as well. Meta holds a 20% membership interest, accounted for under the equity method.

Meta's filing explains why that ownership split matters so much. It says decisions about remarketing the campus, including talks with future tenants and individual property sales, have the most significant impact on the venture's economic performance. Meta says it does not have the power to direct those activities.

That conclusion is the hinge of the whole structure. Because Meta is not the primary beneficiary, it does not consolidate the variable interest entity, and the assets and debt stay off its balance sheet. Meta's auditors had to accept a judgement that the tenant using 100% of the campus does not control the thing that matters most.

Where the bonds actually sit

The second layer of separation is less discussed. IFR notes that the structure splits the issuer of the liabilities, Beignet, from the holder of the asset, JV Co. Its analysis argues that simply consolidating JV Co would not bring the 144A bond onto Meta's balance sheet.

So there are two walls, not one. The first keeps the campus off Meta's books. The second keeps the bond off even if the first wall were ever knocked down. My read is that the second wall is the more deliberate piece of engineering, because it protects the accounting outcome against a future change of view by auditors or standard setters.

Who carries the Hyperion lease obligation

Meta does, and only Meta. It signed operating leases with the joint venture for all of the campus facilities once construction is complete. Each leased property has an initial four-year term, with options to renew for a total period of up to 20 years.

The leases commence in 2029, and Meta's filing puts the aggregate initial lease commitment at approximately $12.31 billion. That figure covers the four-year initial term only. It is less than half of the bond, which is exactly the point.

A four-year lease is unusual for a hyperscale campus. Bisnow's reporting at the time noted that such leases are typically no shorter than 10 years. The short term is what lets Meta argue that renewal beyond year 4 is not reasonably certain, so only 4 years of payments enter the lease accounting.

Why the short term does not make the risk short

IFR describes the leases as four-year non-cancellable blocks with options that can take occupancy to around 20 years. The campus's economic life is put at 25 to 30 years. The bonds amortise to May 2049, roughly 20 years after the leases start.

So the 4-year term is real as a legal matter and misleading as an economic one. A tenant that walks away in 2033 still faces the residual value guarantee, which is designed to make the lenders whole. The option to leave is genuine, but it is priced so that leaving is expensive.

This is a sharper version of the pattern covered in the analysis of $662 billion in uncommenced hyperscaler leases. Those obligations are deferred and will reach the balance sheet at commencement. Hyperion's will mostly not, because the commenced liability is capped at 4 years of rent.

What bondholders can claim if demand falls short

This is the question the deal was built to answer, and the answer depends on when demand falls short. There are three phases, and each has a different protection. The offering memorandum is not public, so the mechanics below come from Meta's filing and IFR's reading of the documents.

During construction

Investors fund construction from day one, with proceeds parked in Treasuries and drawn down as the campus is built. Meta provides construction management. S&P rated this phase one notch below Meta, at A+, because it judged that substantial credit risk passes to Meta during both construction and operation.

During the 16-year guarantee window

If Meta terminates or does not renew a lease, and certain other conditions are met, the residual value guarantee applies. Meta's filing says its maximum payment would equal any shortfall between the property's fair value and the guarantee threshold. The aggregate threshold is approximately $28 billion and decreases over time.

IFR reports that the guarantee also triggers on a Meta default under the lease. It describes the payment as a guaranteed minimum value sized so the bond can be repaid, from the sale of the campus, from Meta directly, or both. In plain terms, if AI demand disappoints and Meta leaves, Meta writes the cheque that the empty buildings cannot.

The final 4 years

The guarantee covers the first 16 years of operations. With leases starting in 2029, that window runs to roughly 2045, while the bonds mature in 2049. IFR reports that those last 4 years are covered by a termination fee equal to the outstanding bond balance at that point.

The gap in what is public is the "certain other conditions" clause. Every protection above has triggers that sit in documents investors saw and the market did not. Anyone treating this as a template should ask for those conditions first, because they decide whether the guarantee is a near-certainty or a contingent promise.

The 144A bond for this data center against a plain Meta bond

The cleanest comparison arrived two weeks later. On 30 October 2025, Meta itself sold $30 billion of senior unsecured notes in six tranches, the largest corporate bond of the year according to IFR. That deal sits on Meta's balance sheet and is a direct claim on the company.

IFR lists the coupons as 4.20% on the 5-year, 4.875% on the 10-year, 5.50% on the 20-year, 5.625% on the 30-year and 5.75% on the 40-year. Beignet paid 6.581% for amortising paper with a final maturity in 2049.

A coupon-to-coupon comparison flatters the gap slightly in both directions. Treasury yields moved between 16 and 30 October, and an amortising bond has a shorter average life than its final maturity suggests. The 20-year tranche at 5.50% is the fairest comparator, which puts the premium near 108 basis points.

The spread at launch tells the same story. IFR reports Beignet priced at Treasuries plus 225 basis points. A later Reuters column put the launch spread at 185 basis points, likely reflecting the price after the bond's early trading pop of about 10 points. I would use IFR's pricing figure, because it is the number investors actually paid.

Why Meta paid extra for off-balance-sheet AI infrastructure debt

On a pure cost basis, Meta could have borrowed the money more cheaply itself. It proved that two weeks later with a record order book. So the extra coupon is the price of something other than cash.

Meta's own framing is flexibility. Its announcement says the four-year term provides long-term strategic flexibility, and that the residual value guarantee balances that optionality in a cost-efficient manner. The 10-Q adds that the venture provides optionality as AI markets and technologies develop.

The accounting result is the other thing being bought. Meta's reported long-term debt was $28.83 billion at 30 September 2025. Adding a $27.3 billion bond would have nearly doubled it in a single quarter, months before capex guidance and the depreciation questions examined in the analysis of GPU useful lives and AI earnings.

My position is that both reasons are real, and that the flexibility argument is weaker than the accounting one. A tenant that must pay a residual guarantee sized to repay the bonds has bought an expensive exit, not a cheap one. The option is worth most in exactly the scenario where exercising it costs the most.

What S&P concluded about leverage

IFR reports that S&P's base case brings in only 8 years of lease payments. It adds that Meta's AA minus rating would be unchanged even if the deal were fully on balance sheet. The binding constraints on Meta's rating are regulatory, not leverage. That cuts against the idea that the structure was needed to protect the rating.

It does not cut against the idea that it protects the optics. Equity investors and many credit models read reported debt first. This is the same visibility gap that drives the private credit exposure to AI debt more broadly: obligations that are economically corporate but legally somewhere else.

What the bond market has said about AI infrastructure debt since

The early verdict was enthusiasm. IFR reports the bonds jumped about 10 points in early trading, and bankers later said they could have achieved a better price. That pop is a sign the deal was priced generously for investors.

The later verdict was less kind. A Reuters column on 29 September 2026 reported the notes trading at 91 cents on the dollar, with a yield as high as 7.55%. It put the spread around 230 basis points, below a record-wide 255 basis points in July 2026.

The same column noted Meta's own credit default swaps had climbed close to 100 basis points. That matters for the risk-transfer question. If the Beignet bonds trade with Meta's credit rather than with the Louisiana real estate, the market has already decided who really holds the risk.

There is a scaling question underneath this. Meta's uncommenced lease obligations reached $278.99 billion at 30 June 2026, against $58.14 billion at 30 September 2025, and it has since set up a second venture in El Paso. Hyperion is the template, not the exception, which makes the payback math on AI capex a question for bondholders as well as shareholders.

Where this argument is weakest

Three points cut against the risk-transfer critique, and they are not trivial.

First, nothing is hidden. Meta announced the deal, named the parties and states its maximum exposure in every quarterly filing. IFR's own headline compares the structure with Enron, but Enron's problem was undisclosed obligations. A reader who opens Note 5 has the number, and the comparison flatters the critics more than the facts do.

Second, the rating agency looked through it. If S&P would keep Meta at AA minus with the whole deal consolidated, the structure is not concealing credit weakness. It is changing presentation. Presentation matters, but it is a smaller charge than misrepresentation.

Third, Blue Owl's equity is genuinely at risk. If the guarantee conditions are not met, the first loss falls on Blue Owl's funds, not on Meta. Our critique leans on IFR's reading of an offering memorandum we have not seen. If those conditions are narrow, the risk transfer is more real than this post suggests.

The evidence base has its own weak spot. Secondary trading figures come from a single column, and the RVG trigger conditions are not public. A weaker point still is our premium estimate, which compares bonds priced two weeks apart. I would not lean on the 108 basis point figure beyond its first digit.

Frequently asked questions

Where to start on the next 144A data center deal

Hyperion will not be the last of these, and the next one will be announced with a press release that leads on ownership. Read it backwards.

Start with the tenant's quarterly filing, not the announcement. Find the equity method or VIE note and write down two numbers: the carrying value of the investment and the maximum exposure to loss. The ratio between them is the single best summary of how much risk actually moved. For Hyperion it is roughly 16 to 1.

Then ask one question of anyone selling you the bonds or the thesis: what are the conditions on the residual value guarantee? If nobody can answer, you are pricing Meta credit with project bond documentation, and you should demand the spread of the weaker of the two. The same discipline applies to neocloud debt stacks, where the tenant is often less creditworthy than Meta.

Related reading