Wiki/Primary documents

Hyperion SPV ($27.3B), Annotated

Source set: Meta press release (Oct 21, 2025); Beignet Investor LLC 144A offering (priced Oct 16, 2025); S&P Global Ratings on Beignet Investor LLC; Meta investor materials; Blue Owl Capital disclosures; IFR Awards write-up of the deal; Bloomberg, Fortune, Data Center Dynamics, Bisnow, Data Center Frontier reporting. Pulled May 2026.

Why this deal matters

There are now roughly a dozen named, ≥1 GW AI campuses under construction in the U.S. There is exactly one for which a third-party rating agency has published a full credit opinion, a 144A/Reg S note has cleared the market, and the operating-tenant economics are visible from outside the SPV.

That deal is Beignet Investor LLC — the bankruptcy-remote SPV issuing vehicle that financed the Meta–Blue Owl Hyperion campus in Richland Parish, Louisiana. On October 16, 2025, Beignet priced US$27.294 billion of 6.581% senior secured notes due May 30, 2049 at T+225 bp, rated A+ by S&P (single-agency), in 144A-for-life format. Five days later, on October 21, Meta and Blue Owl jointly announced the JV publicly.

Every other gigawatt-scale AI campus deal disclosed to date is either (a) wholly on a hyperscaler's balance sheet, (b) inside a private REIT structure (QTS, Aligned) where the operating economics are opaque, or (c) at an earlier-stage equity partnership without the debt-tranche disclosure to anchor the analysis (GAIIP, CoreWeave's Crusoe lease). Hyperion is the cleanest disclosed deal-economics datapoint we have on how the financial system is intermediating ≥1 GW AI infrastructure.

It is also the largest private-credit transaction in history by some margin — the prior record was the ~$15 billion Inflation Reduction Act-linked bridge for First Brands in 2024 — and the largest single project-financing of any kind ever booked as an investment-grade bond. That those two records were broken simultaneously, on one deal, on one campus, in rural Louisiana, is the story.

The shape is now being copied. Meta is reportedly assembling a ~$13B SPV for a Texas campus on a similar structure; Aligned/BlackRock, QTS/Blackstone, and the broader AIP (formerly GAIIP) consortium are all working in this template. If you only read one financing in the AI capex cycle, read Beignet.

The capital stack

From Meta's October 21, 2025 press release (investor.atmeta.com, mirrored on about.fb.com):

"Funds managed by Blue Owl Capital will own an 80% interest in the joint venture, while Meta will retain the remaining 20% ownership. The parties have committed to fund their respective pro rata share of the approximately $27 billion in total development costs for the buildings and long-lived power, cooling, and connectivity infrastructure at the campus."

"Blue Owl Capital made a cash contribution of approximately $7 billion to the joint venture, and Meta received a one-time distribution from the joint venture in the amount of approximately $3 billion."

"A portion of capital raised by Blue Owl will be funded by debt issued to PIMCO and select other bond investors through a private securities offering. Morgan Stanley & Co. LLC served as exclusive financial advisor to Meta in connection with this transaction and served as sole bookrunner in connection with the private securities offering."

Translated to a stack — combining the press release with IFR's deal write-up and S&P's pre-sale on Beignet:

Tranche	Amount	Holder	Terms
Equity (80%)	~$7.0B cash (plus deferred)	Blue Owl-managed funds	80% JV LP interest
Equity (20%)	~$2.5B in-kind (land, predevelopment)	Meta Platforms	20% JV LP interest; $3B one-time distribution received
Senior secured notes	$27.294B	PIMCO (~$18B anchor), BlackRock (>$3B), small handful of others	6.581% coupon, fully amortizing, due May 30, 2049, A+ (S&P), 144A/Reg S, T+225 bp at pricing

Three structural points are worth pulling out:

1. The debt is fully amortizing, single-tranche, 23.6-year WAL. This is extraordinarily rare in investment-grade markets, which favor bullet maturities. Per IFR Awards, a Morgan Stanley banker described it as "a low double-digit number of investors in a deal of US$27bn" — i.e., average ticket north of $1B. PIMCO took ~$18B; BlackRock took "more than $3 billion" (Bloomberg / Fortune). One unidentified buyer called the amortizing profile "an extremely attractive feature," noting that principal returns to investors "even if AI/data centres have issues in the future" — a tell that buyers are pricing some technology- obsolescence risk despite the A+ stamp.

2. The equity is thin: ~$9.5B of equity (Meta + Blue Owl) against $27.3B of debt is 22% equity / 78% debt at completion, or about ~8.5% equity if you count only the cash contribution against gross project cost. For comparison, a typical merchant-power project bond clears at ~30–40% equity. The thin cushion is what S&P is reaching through Meta to underwrite (see "Rating agency assessments" below).

3. The "$3 billion one-time distribution" to Meta is the part the mainstream coverage missed. Meta contributes land and predevelopment work — call it $2–3B of cost basis — and receives $3B back at closing from Blue Owl's equity. The arithmetic implies Meta is exiting its upfront cost on the deal and starting the build with a positive cash position on the 20% JV stake, while still retaining operational control, the lease, and the residual value guarantee (RVG). For a transaction booked as off-balance-sheet, the cash mechanics are net-inflow to Meta at signing.

The lease structure: why a 4-year term against a 30-year-life asset is unusual

This is the single most interesting feature of Hyperion and the one the press release describes most carefully:

"Meta entered into operating lease agreements with the joint venture for use of all of the facilities of the campus once construction is complete, with these lease agreements having a four-year initial term with options to extend."

"Meta provided the joint venture with a residual value guarantee for the first 16 years of operations whereby Meta would make a capped cash payment to the joint venture based on the then-current value of the campus if certain conditions are met following a non-renewal or termination of a lease."

A 4-year initial term against a steel-and-concrete asset with a 30-year physical life is not a coincidence; it is the entire accounting trick.

Under ASC 842 (operating-lease accounting), a lessee must consolidate the leased asset on its balance sheet if any of the following are true: the lease transfers ownership; the lease contains a purchase option reasonably certain to be exercised; the lease term is for the major part of the asset's economic life; or the present value of lease payments equals substantially all of the asset's fair value. A 4-year term on a 30-year asset clears all of those tests with room to spare.

The "options to extend" are critical. Per Michael Cembalest's analysis (JPMorgan; published Nov 2025 on LinkedIn) and the satirical-but- sourced Stohl Substack reconstruction of the S&P pre-sale, Meta is the lessee through a sub-entity (reported as "Pelican Leap LLC") under what is described as eleven triple-net leases — one per building — with a 4-year initial term beginning at 2029 completion, four renewal options, and the structure sized to potentially extend to ~20 years if all options are exercised. For Meta to keep the campus off-balance-sheet, it must assert at inception that renewal is "not reasonably certain" — despite having designed, partly funded, and operationally controlled a 2+ GW campus purpose-built for its AI training and inference workloads.

The residual value guarantee is what makes that assertion economically real. Per Cembalest, the RVG declines over time, remains higher than outstanding debt at every measurement date, and reaches zero after 20 years. If Meta doesn't renew at year 4 (or any subsequent option date), the SPV sells the campus; if sale proceeds fall short of the guaranteed residual value, Meta cuts a "capped cash payment" to the JV to make bondholders whole. S&P's own commentary (as reported in the deal docs and reconstructed in secondary coverage) is that the agency excludes the RVG from Meta's adjusted leverage on the grounds that it is contingent — and only triggered in a non-renewal-plus-impaired- sale scenario.

This is the cleverest part of the deal. From the debt investor's perspective the credit is "Meta backstops the bonds." From Meta's accounting perspective, neither the campus, the $27.3B of debt, nor the RVG sits on its balance sheet, and only the present value of the four years of contracted rent (plus eight years of assumed renewal, per S&P's adjusted-debt methodology) shows up. Cembalest estimates this adds roughly 0.2x to Meta's adjusted leverage at consolidation, versus the ~3x it would add if the full $27.3B were on-balance-sheet.

For a 30-year asset financed with a 23.6-year amortizing bond serviced by a 4-year cancelable lease, the construction can be summarized: the lease term is a accounting fiction; the RVG is the actual economic commitment. Meta is on the hook for 16 years.

What Meta disclosed in its 10-K and earnings

Important caveat: Meta's FY2025 10-K had not been filed at the time the Hyperion JV was announced (October 21, 2025) — the JV signing happened in the middle of Q4 2025, so the disclosures landed in the Q3 2025 10-Q (filed Oct 29, 2025) and then the FY25 10-K filed in late January 2026. We have not pulled the 10-K filing text directly in this writeup; the language quoted below is from the Q3 2025 10-Q disclosure that the JV announcement and secondary coverage references. Verbatim 10-K disclosure should be sourced from EDGAR (CIK 0001326801) when refreshing this page.

From the Q3 2025 10-Q (per Meta's own press release language and Bisnow reporting):

"Meta entered into operating lease agreements with the joint venture for use of all of the facilities of the campus once construction is complete, with these lease agreements having a four-year initial term with options to extend. Meta also provided the joint venture with a residual value guarantee for the first 16 years of operations whereby Meta would make a capped cash payment to the joint venture based on the then-current value of the campus if certain conditions are met following a non-renewal or termination of a lease."

From Susan Li, CFO, on Meta's Q2 2025 earnings call (the call before the Hyperion JV was named publicly, but where the structural template was previewed):

"We generally believe that there will be models here that will attract significant external financing to support large-scale data center projects that are developed using our ability to build world-class infrastructure while providing us with flexibility should our infrastructure requirements change over time."

The phrase "flexibility should our infrastructure requirements change over time" is the load-bearing one. It is the management-narrative version of "we don't have to consolidate this if we don't extend." It is also a tell on tenant-side optionality value: Meta is paying for the right to walk away at year 4 (or 8, 12, 16) and that optionality is priced into the rent.

What Meta has not disclosed publicly (as of May 2026):

The contracted rent per kW or per square foot.
The implied yield-on-cost from the SPV's perspective.
The precise mechanics of the RVG cap — only that there is one.
Whether the renewal options carry a step-up in rent (likely yes).
Whether other Meta SPVs (the rumored Texas $13B deal) use identical terms.

What Blue Owl disclosed in its filings

Blue Owl's disclosure shape is the inverse of Meta's: where Meta is deliberately understating presence in the JV, Blue Owl is loudly marketing it. From the Co-CEOs' joint statement in the press release:

"We're proud that our funds are partnering with Meta on the development of the Hyperion data center campus — an ambitious project that reflects the scale and speed required to power the next generation of AI infrastructure." — Doug Ostrover and Marc Lipschultz, Blue Owl Capital Co-CEOs

The capital deployed is split across several Blue Owl-managed funds — the firm has not broken the $7B cash contribution down by fund in publicly available disclosure, but its Digital Infrastructure strategy (formerly IPI Partners, acquired April 2024) is the obvious home for the bulk of it. Kirkland & Ellis served as Blue Owl's counsel; Morgan Stanley was sole bookrunner on the debt and exclusive advisor to Meta. Per Cembalest's Mea Culpa LinkedIn post, Blue Owl also cites a minimum IRR guarantee that Meta is responsible for meeting — i.e., on the equity side, Blue Owl's 80% LP economics are underwritten not just by the lease cash flow and RVG, but by an incremental return guarantee from Meta. This is not in the press release. If it holds up, it pushes the deal closer to "synthetic secured lending dressed up as equity" than to a real equity partnership.

Blue Owl's 8-K and Q4 2025 earnings call did flag Hyperion as a flagship transaction for its Digital Infrastructure platform. Yahoo Finance and PE Insights both report the deal "redefined Blue Owl's investment story," and the firm has used it as the marquee reference in 2026 fundraising.

Rating agency assessments

S&P Global Ratings is the only agency to have rated the Beignet notes. Moody's and Fitch did not assign ratings, which is itself a data point: a $27.3B IG deal is typically rated by at least two agencies to satisfy investor mandate constraints. PIMCO's willingness to anchor at $18B on a single rating signals (a) the anchor was negotiated long before the bond marketed and (b) Meta provided sufficient direct diligence that PIMCO's internal credit work substituted for a second public rating.

S&P's rating decision, summarized from the agency's public regulatory disclosure (spglobal.com/ratings) and the IFR Awards write-up:

Rating: A+ (stable).
The rating is "notched off" Meta's senior unsecured rating (Aa3/AA– per Moody's/S&P) with a one-to-two-notch differential reflecting structural subordination through the SPV.
The credit story rests on "very strong contractual linkage to Meta" and the residual value guarantee through the first 16 years of operations.
S&P consolidates only 8 years of assumed lease payments into Meta's adjusted debt at lease commencement in 2029 — not the full potential lease term, not the RVG.
The DSCR (debt service coverage ratio) on the bonds is approximately 1.12x through 2049 under the contracted lease (per the Stohl reconstruction of the pre-sale).

A 1.12x DSCR is thin — merchant power project bonds typically need 1.4–1.5x. The structure clears at 1.12 only because the contracted lease + RVG combination effectively translates Meta's IG balance sheet into the cash flow that services the bonds. Strip out the Meta backstop and the deal is sub-investment-grade. That is the most important thing to know about Hyperion.

The financing economics implied

The deal pricing lets us back into rough yield-on-cost economics from the SPV's perspective. Working with public numbers only, treating ranges where exact terms aren't disclosed:

Bond service: $27.294B at 6.581% on a fully amortizing 23.6-year profile. First-year cash interest is ~$1.80B; level annual debt service (P+I) on a 23.6-year amortization is approximately $2.30–2.40B per year.

Equity required return: Blue Owl's 80% equity ($7B cash) underwriting infrastructure-fund target IRRs of 10–12% (per Blue Owl's public disclosure on its Digital Infrastructure strategy), with downside protection from Meta's reported minimum-IRR guarantee. Annual yield target to Blue Owl: roughly $0.7–0.9B.

Total annual cash flow required to service capital structure (debt

Blue Owl equity return): ~$3.0–3.3B per year.

Implied gross rent on a $27B project: ~11–12% of gross development cost. That maps tightly to Digital Realty's disclosed Q4 2025 expected stabilized yield-on-cost of 11.9% on $10B / 769 MW under construction — the most reliable comparable in the listed-REIT universe. It is not the 25–30% gross-revenue yield in the Sacks math.

In other words: the price the credit market is willing to pay PIMCO/ BlackRock to fund Hyperion implies that Meta is paying rent at roughly the same yield-on-cost an experienced colocation REIT underwrites to — about 11–12%. The "AI exceptionalism" priced into hyperscaler equity does not appear to be priced into the project-level debt. Beignet bondholders are getting paid roughly the same risk spread (T+225) that a high-grade utility or REIT bond would clear at in the current rate environment.

(One caveat: 6.58% IG money in October 2025 was, in fact, closer to high-yield territory. Fortune's coverage explicitly flagged this: "a level closer to high-yield, or 'junk,' bond territory." The absolute coupon reflects the back-end rate move in 2025; the spread of 225 bp is the IG signal.)

Comparables

Deal	Date	Size	Structure	What it tells us
GAIIP / AIP (Microsoft + BlackRock + GIP + MGX, later joined by Nvidia + xAI)	Sep 2024	$30B equity → ~$100B with debt	Multi-LP equity platform; deploys into individual deals	The fundraising vehicle. AIP is the equity side of the same template — Hyperion is the asset side.
Aligned / AIP+MGX/GIP	Oct 2025	$40B enterprise value	Private take-private of a 5 GW operator from Macquarie	The platform acquisition. Same investor consortium (BlackRock/GIP/MGX/Microsoft/Nvidia) as GAIIP, now buying scaled colocation rather than greenfielding.
Beignet / Hyperion	Oct 2025	$27.3B debt + $9.5B equity	Bankruptcy-remote SPV, single hyperscaler anchor tenant, ABS-like amortizing bond	The single-asset, single-tenant project bond. The cleanest disclosure.
QTS / Blackstone (CMBS + multi-tranche)	Multiple 2024–26	$3.5B CMBS refi, $1.95B CMBS, $1.65B private bonds, $4.6B planned green bond, $12.5B peak demand on debut HG sale	Multi-asset CMBS plus IG corporate debt	The platform-level debt model. Compare to Beignet's single-asset project bond — Blackstone aggregates risk across the QTS portfolio; Beignet isolates it.
Blackstone Pennsylvania	Jul 2025	$25B announced + $60B catalyzed	Mixed digital + power infrastructure equity	The state-level "pipeline" pre-commitment. No bond yet.
Meta Texas SPV (rumored)	Reported Q1 2026	~$13B	Same template as Beignet, smaller campus	If executed on identical terms, validates that Beignet is a repeatable structure rather than a bespoke one.

The cleanest read across these: GAIIP is the equity vehicle, Aligned is the platform acquisition, QTS is the multi-asset securitization play, and Beignet is the single-asset project bond. All four structures are now in market simultaneously. Beignet is the most informative because its single-asset, single-tenant nature forces the deal economics to be transparent rather than blended.

Open questions

What happens at year 4 if Meta doesn't extend? The press release says the RVG is "capped" — capped at what? If the cap is below the outstanding debt balance plus required Blue Owl equity return, the structure has a real tail risk: a non-renewal in 2033 leaves the SPV with a single-purpose AI campus, a depleted GPU fleet, and amortized debt against a stranded asset in rural Louisiana. The RVG cap level is the single most important number not in the press release.
What is the per-kW contracted rent? Meta hasn't disclosed it. The implied 11–12% yield-on-cost backs into a rent figure — at 2.1 GW IT capacity and 2029 commencement, roughly $130–145/kW-month — but that number is reconstructed, not disclosed.
Does the RVG decline schedule match the bond amortization schedule? Cembalest reports the RVG "remains higher than outstanding debt at every measurement date." If true, the RVG functions as a perfect bondholder backstop through 2045 and the "operating lease" framing is purely an accounting overlay. If the RVG ever drops below outstanding principal — even by $1 — the credit story changes.
Why no second rating? $27.3B with one agency is unusual. Either Meta declined a second rating (saves fees, avoids unfavorable notch) or one of Moody's/Fitch was approached and declined to rate the structure. Either answer is informative.
What's in the data room? The 144A-for-life format means the offering memorandum is only accessible to qualified institutional buyers under NDA. The full lease, the full RVG schedule, the covenant package, and the SPV's voting rights vs. Meta's tenant rights are all there, and none of it is public. The press release discloses the shape of the deal; the OM discloses the terms.
Is the structure replicable at smaller scale, or does it require ≥$20B size to justify the legal complexity? If Meta's rumored $13B Texas SPV prices on equivalent terms, the answer is "yes, replicable at $10B+." If the Texas deal prices wider (or doesn't get rated A+), the Hyperion structure may be unique to a single confluence of size, tenant credit, and 2025 anchor-investor appetite.
What happens to the structure if Meta's IG rating notches down? The A+ on Beignet is notched from Meta's senior unsecured. A one-notch downgrade at Meta likely drags Beignet to A. A multi-notch downgrade — say, into the BBB range, which is where the credit market's AI bears (and Meta's widening CDS, per ZeroHedge in October 2025) imply the equity-market-implied rating already is — would trigger covenant tests inside the bond docs that we cannot see from outside.

Sources

Primary disclosures

Meta press release, October 21, 2025 — investor.atmeta.com ; about.fb.com
S&P Global Ratings on Beignet Investor LLC — spglobal.com/ratings (regulatory disclosure)
Meta Q3 2025 10-Q (CIK 0001326801) — EDGAR; filed Oct 29, 2025
Meta FY2025 10-K — EDGAR; filed late January 2026 (verbatim disclosure to be pulled on refresh)
Kirkland & Ellis client release (deal counsel to Blue Owl) — kirkland.com

Deal-press coverage (secondary)

IFR Awards 2025 — Financing Package: Blue Owl Capital / Beignet Investor's US$27.3bn 23.6-year bond — ifre.com
Bloomberg, "Meta, Blue Owl Seal $30 Billion Private Capital Deal for AI" — Oct 16, 2025
Fortune, "Meta's $27 billion bet turns AI compute into Wall Street's hottest new investment" — Oct 31, 2025
Bisnow, "Meta Pushes Its Largest Data Center Project Off Its Books With $27B JV" — bisnow.com
Data Center Dynamics, "Meta forms $27bn joint venture with Blue Owl" — datacenterdynamics.com
Data Center Frontier, "Meta's Dual-Track Data Center Strategy" — datacenterfrontier.com
Proximo, "Meta and Blue Owl's Hyperion: Beyond project bonds" — proximoinfra.com (paywall)

Analyst commentary

Michael Cembalest (JPMorgan), "Mea Culpa: my comments on Blue Owl SPV economics with Meta" — November 2025, LinkedIn
TheValueist on X, "$META $OWL executive summary" thread — x.com/TheValueist/status/1984231167290139044
Stohl Substack reconstruction of S&P pre-sale — stohl.substack.com (treat as a reconstruction, not the agency's own text)
Ernest Chiang, "Off-Balance Sheet AI: How SPVs Are Financing the Data Center Boom" — ernestchiang.com
Global Data Center Hub, "Meta + Blue Owl's $27B Bet" — globaldatacenterhub.com

Comparables

BlackRock/GIP/Microsoft/MGX GAIIP launch press release — September 17, 2024 — news.microsoft.com
AIP / MGX / GIP acquisition of Aligned Data Centers — October 2025 — global-infra.com
Blackstone $25B Pennsylvania announcement — July 2025 — blackstone.com
QTS multi-tranche debt program (CMBS + IG corporate) — Goldman Sachs–led $2B loan and successive 2024–2026 issuance — costar.com

Louisiana state and utility filings

Louisiana Economic Development announcements — opportunitylouisiana.gov (Meta Richland Parish project; not pulled verbatim in this writeup)
Entergy Louisiana LPSC docket (three gas turbines, ~2.26 GW combined, approved 2025; subsequently expanded to ten plants / 7.5 GW per March 2026 announcement) — datacenterdynamics.com
Meta, "One Year In: Meta's Richland Parish Data Center Supports Louisiana Economy With $875M In Contracts" — December 2025 — about.fb.com

Verification notes: The SPV name "Beignet Investor LLC" is confirmed from multiple secondary sources (Fortune, IFR, TheValueist) and from cbonds.com's issuer record. The "Pelican Leap LLC" tenant entity name comes from a single reconstructed source (Stohl Substack) and should be treated as unverified until the actual offering memorandum or a direct UCC filing is pulled. The 6.581% coupon, $27.294B principal, May 30, 2049 maturity, A+ S&P rating, T+225 bp spread, and PIMCO $18B / BlackRock $3B+ anchor sizing are all corroborated across at least three independent secondary sources. The 11–12% implied yield-on- cost is a reconstruction from disclosed coupon and structure, not a disclosed figure. Refresh on next 10-K and any subsequent S&P credit update for Beignet.