Meta 利用高级几何隐藏 270 亿美元债务
Meta hiding $27B in debt using advanced geometry

原始链接: https://stohl.substack.com/p/exclusive-credit-report-shows-meta

## Beignet Investor LLC – 初步A+评级摘要 FSG LLC 为 Beignet Investor LLC 的 273 亿美元高级担保票据分配了初步 A+ 评级,该票据与 Meta Platforms (AA-/稳定) 的 80% 合资企业相关,用于在路易斯安那州建设 287.9 亿美元的超大规模数据中心园区。该结构允许 Meta 为项目提供资金和控制,同时使其相关的债务大部分不计入其资产负债表。 评级取决于 Meta 强大的合同义务——涵盖建设超支、租赁付款和剩余价值保证——有效地将所有重大风险 *转移* 给 Meta。预计现金流稳定,独特的融资机制利用借款利息来满足股权要求。 FSG 承认内在风险——对单一租户的集中度、潜在的会计变更以及对有利市场条件的依赖——但由于 Meta 的信用状况,可以容忍这些风险。展望为“表面稳定”,意味着如果出现意外问题,评级可能会被下调。本质上,该评级反映了人们的信念,即 Meta 将有效地为整个项目提供支持,尽管该结构的设计旨在 *显得* 独立。

## Meta 270 亿美元债务通过复杂金融结构隐藏 一份最新报告详细说明了 Meta 如何在路易斯安那州建设一个价值 280-290 亿美元的人工智能数据中心,同时似乎避免在资产负债表上增加 273 亿美元的债务。他们通过一个有限责任公司网络,主要通过 Beignet Investor LLC 公司借入资金来实现这一目的。 尽管 Meta 设计了园区,承担超支成本,保证其价值,并且是唯一的用户,但由于会计规则的技术性问题,这笔债务在法律上被归类为“别人的”。这些规则规定,只有当公司“控制”该实体并承担风险时,债务才会出现在资产负债表上。Meta 辩称它不控制该项目,尽管事实证明它确实如此。 评级机构正在嘲笑这种安排,他们承认这种荒谬性,但受当前会计解释的约束。本质上,Meta 在借钱建设自己的数据中心,但将其结构化为外部义务,引发了人们对这种做法的普遍性以及财务透明度的精神的质疑。
相关文章

原文

28-Nov-2025 | 09:12 CST

FSG LLC (Flexible Standards Group)

Analyst: Tom Bellwether
Contact Information: None available*
*Because of the complex nature of financial alchemy, our analysts live a hermetic lifestyle and avoid relevant news, daylight, and the olfactory senses needed to detect bullshit.

Following our review of Beignet Investor LLC (the Issuer), an affiliate of Blue Owl Capital, in connection with its participation in an 80% joint venture with Meta Platforms Inc., we assign a preliminary A+ rating to the Issuer’s proposed $27.30 billion senior secured amortizing notes.

This rating reflects our opinion that:

  1. All material risks are contractually assigned to Meta, which allows us to classify them as hypothetical and proceed accordingly.

  2. Projected cash flows are sufficiently flat and unbothered by reality to support the rating.

  3. Residual Value Guarantees (RVGs) exist, which we take as evidence that asset values will behave in accordance with wishes rather than markets.

The Outlook is Superficially Stable, defined here as “By outward appearances stable unless, you know, things happen. Then we’ll downgrade after the shit hits the fan.”

Blue Owl Capital Inc. (Blue Owl, BBB/Stable), through affiliated funds, has created Beignet Investor LLC (Beignet or Issuer), a project finance-style holding company that will own an 80 percent interest in a joint venture (JVCo) with Meta Platforms Inc. (Meta, AA-/Stable). The entity is named “Beignet,” presumably because “Off-Balance-Sheet Leverage Vehicle No. 5” tested poorly with focus groups.

Beignet is issuing $27.30 billion of senior secured amortizing notes due May 2049 under a Rule 144A structure.

Note proceeds, together with $2.45 billion of deferred equity from Blue Owl funds and $1.16 billion of interest earned on borrowed money held in Treasuries, will fund Beignet’s $23.03 billion contribution to JVCo for the 2.064 GW hyperscale data center campus in Richland Parish, Louisiana, along with reserve accounts, capitalized interest and other transaction costs that seem small only in comparison to the rest of the sentence.

Iris Crossing LLC, an indirect Meta subsidiary, will own the remaining 20 percent of JVCo and fund approximately $5.76 billion of construction costs.

We assign a preliminary A+ rating to the notes, one notch below Meta’s issuer credit rating, reflecting the very strong contractual linkage to Meta and the tight technical separation that allows Meta to keep roughly $27 billion of assets and debt off its balance sheet while continuing to provide all material economic support.

Meta transferred the Hyperion data center project into JVCo, which is owned 80 percent by Beignet and 20 percent by Iris Crossing LLC, an indirect Meta subsidiary. JVCo, in turn, owns Laidley LLC (Landlord). None of this is unusual except for the part where Meta designs, builds, guarantees, operates, funds the overruns, pays the rent, and does not consolidate it.

This project has nine data centers and two support buildings, with about four million sq. ft. and 2.064 GW capacity. The support buildings will store the reams of documentation needed to convince everyone this structure isn’t what it looks like. The total capital plan of $28.79 billion will be funded as follows:

The structure allows the Issuer to borrow money, earn interest on the borrowed money, and then use that interest to satisfy the equity requirement that would normally require… money.

Nothing is created. Nothing is contributed. It’s a loop. Borrow money, earn interest, and use the interest to claim you provided equity. The kind of circle only finance can call a straight line.

Together, these flows cover Beignet’s $23.03 billion obligation to JVCo, plus the usual constellation of capitalized interest, reserve accounts, and transaction expenses. In any other context this would raise questions. For us, it raises the credit rating.

Meta, through Pelican Leap LLC (Tenant), has entered into eleven triple-net leases—one for each building—with an initial four-year term starting in 2029 and four renewal options that could extend the arrangement to twenty years. The leases rely on the assumption that Meta will continue to need exponentially more compute power and that AI demand will not collapse, reverse, plateau, or become structurally inconvenient.

The notes issued by Beignet are secured by Beignet’s equity interest in JVCo and relevant transaction accounts. They are not secured by the underlying physical assets, which remain at the JVCo and Landlord level. This is described as standard practice, which is true in the same way that using eleven entities to rent buildings to yourself has become standard practice.

The resulting structure allows Meta to support the project economically while leaving the associated debt somewhere that is technically not on Meta’s balance sheet. The distinction is thin, but apparently wide enough to matter.

The preliminary A+ rating reflects our view that this is functionally Meta borrowing $27.30 billion for a campus no one else will touch, packaged in legal formality precise enough to satisfy the letter of consolidation rules and absurd enough to insult the spirit.

Credit risk aligns almost one-for-one with Meta’s own profile because:

Taken together, these features tie Beignet’s credit quality to Meta so tightly that you’d have to not be paying attention to miss them. The structure maintains a precarious technical separation that, under current interpretations of accounting guidance, allows Meta to keep roughly $27 billion of assets and debt off its own balance sheet while continuing to provide every meaningful form of economic support.

This treatment is considered acceptable because the people who decide what is acceptable have accepted it.

Share

JVCo qualifies as a variable interest entity because the equity at risk is ceremonial and the real economic exposure sits entirely with the party insisting it does not control the venture. This remains legal due to the enduring belief that balance sheets are healthier when the risky parts are hidden.

Under U.S. GAAP, consolidation is required if Meta is the primary beneficiary, defined as the party that both:

  1. Directs the activities that most significantly affect the entity’s performance, and

  2. Absorbs significant losses or receives significant benefits.

Meta asserts it is not the primary beneficiary.

To evaluate that assertion, we note the following uncontested facts:

  • Meta is responsible for designing, overseeing, and operating a 2.064 GW AI campus, an activity that requires technical capabilities Blue Owl does not possess.

  • Meta bears construction cost overruns beyond 105 percent of the fixed budget, as well as specified casualty repair obligations of up to $3.125 billion per event during construction.

  • Meta provides the guarantee for all rent and operating payments under the leases, across the initial term and any renewals.

  • Meta provides the residual value guarantee, ensuring bondholders are repaid if leases are not renewed or are terminated, either through a sale or by paying the guaranteed minimum values directly.

  • Meta contributes funding, directs operations, bears construction risk, guarantees payments, guarantees asset values, determines utilization, controls renewal behavior, and can trigger the sale of the facility.

Based on this, or despite this, Meta concludes it does not control JVCo.

Our interpretation is fully compliant with U.S. GAAP, which prioritizes the geometry of the legal structure over the inconvenience of economic substance and recognizes control only if the controlling party agrees to be recognized as controlling.

Meta has not agreed, and the framework, including this agency, respects that choice.

For rating purposes, we therefore accept Meta’s non-consolidation as an accounting outcome while treating Meta, in all practical respects, as fully responsible for the performance of an entity it does not officially control.

The lease structure is designed to look like a normal commercial arrangement while functioning as a long-term commitment Meta insists, for accounting reasons, it cannot possibly predict.

Tenant will pay fixed rent for the first 19 months of operations, based on a 50 percent assumed utilization rate, after which rent scales with actual power consumption. The leases are triple-net. Meta is responsible for everything: operating costs, maintenance, taxes, insurance, utilities. If a pipe breaks, Meta fixes the pipe. If a hurricane relocates a roof, Meta pays to staple the roof back on.

In practical terms, the only scenario in which Beignet bears operating exposure is a scenario in which Meta stops paying its own bills, at which point the lease structure becomes irrelevant because the same lawyers that structured this deal will have already quietly extricated Meta from liability.

The contract terms include:

  • A minimum rent floor engineered to produce a DSCR of 1.12 in a spreadsheet where 1.12 was likely hard-coded and independent of math.

  • A four-year initial term with four four-year renewal options, theoretically creating a 20-year runway Meta pretends not to see.

  • Meta guarantees all tenant payment obligations across the entire potential lease life, including renewals it strategically refuses to acknowledge as inevitable.

  • No performance-based KPIs. Under this structure, the buildings could underperform, overperform, or catch fire. Meta still pays rent.

  • The RVG requires Meta to ensure that, at every potential lease-termination date, the asset is worth at least the guaranteed minimum value. If markets disagree, Meta pays the difference. Because Meta is rated AA-/Stable, we are instructed to assume that it will do so without hesitation, including in scenarios where demand softens or secondary markets discover that a hyperscale campus in Richland Parish is not the world’s most liquid asset class.

The interplay between the lease term and the RVG creates a circular logic we find structurally exquisite.

From a credit perspective, this circularity is considered supportive, because the same logic used to avoid consolidating the debt also ensures bondholders are paid. The circularity is not treated as a feature or a flaw. It is treated as accounting.

Because Meta is AA-/Stable, we assume it will pay whatever number the Excel model finds through Goal Seek, even in scenarios involving technological obsolescence or an invasion of raccoons.

The accounting hinges on a paradox engineered with dull tweezers:

  • Under lease accounting, Meta must record future lease obligations only if renewals are reasonably certain.

  • Under RVG accounting, Meta must record a guarantee liability only if payment is probable.

To keep $27 billion off its balance sheet, Meta must therefore assert:

  • Renewals are not reasonably certain, despite designing, funding, building, and exclusively using a 2.064 GW AI campus for which the realistic tenant list begins and ends with Meta.

  • The RVG will probably never be triggered, despite the fact that not renewing would trigger it immediately.

This requires a narrow corridor of assumptions in which Meta simultaneously plans to use the facility for two decades and insists that no one can predict four years of corporate intention.

From a credit standpoint, we are supportive. The assumptions that render the debt invisible are precisely what make it secure. A harmony best described as collateralized cognitive dissonance.

  • Meta linkage. The economics are wedded to Meta’s credit profile, which we are required to describe as AA-/Stable rather than “the only reason this entire structure doesn’t fold from a stiff breeze.” Meta guarantees the rent, the RVG, and the continued relevance of the facility. The rest is décor auditors would deem “tasteful.”

  • Minimum rent floor. The lease schedule produces a perfectly flat DSCR of 1.12 through 2049. Projects of this size do not produce flat anything, but the model insists otherwise, so we pretend we believe it. Being sticklers for tradition, and having learned nothing from the financial crisis of 2008, we treat the spreadsheet as the final arbiter of truth, even when the inputs describe a world no one lives in.

  • Construction risk transfer. Meta absorbs cost overruns beyond 105 percent of budget and handles casualty repairs during construction. Our methodology interprets “contractually transferred” as “ceased to exist,” so we decline to model the risk of overruns on a $28 billion campus built in a hurricane corridor. This is considered best practice.

  • RVG backstop. The residual value guarantee eliminates tail risk in much the same way a parent cosigning for their teenager’s car loan eliminates tail risk: by ensuring that the person with all the money pays for everything. If the market value collapses, Meta pays the difference. If the facility can’t be sold, Meta pays the whole thing. If the entire campus becomes a raccoon sanctuary, Meta still pays. We classify this as credit protection, a nuanced designation that allows us to recognize the security of the arrangement without recognizing the debt.

  • Absence of performance KPIs. There are no operational KPIs that allow rent abatement. This is helpful because KPIs create volatility, and volatility requires thought, a variable we explicitly exclude from our methodology. By removing KPIs entirely, the structure ensures a level of cash-flow stability that exists only in transactions where the tenant is also the economic owner pretending to be a squatter.

The rating also reflects several risks that are acknowledged, intellectually troubling, and ultimately tolerated because Meta is large enough that everyone agrees to stop asking questions.

  • Off-balance-sheet dependence. Meta treats JVCo as if it belongs to someone else, which is a generous interpretation of ownership. If consolidation rules ever evolve to reflect economic substance, Meta could be required to add $27 billion of assets and matching debt back onto its own balance sheet. Our methodology treats this as a theoretical inconvenience rather than a credit event, because calling it what it really is would create a conflict with the very companies we rate.

  • Concentration risk. The entire project exists for one tenant with one business model in one industry undergoing technological whiplash. The facility is engineered so specifically for Meta’s AI ambitions that the only plausible alternative tenant is another version of Meta from a parallel timeline. We strongly disagree with the many-worlds interpretation of quantum mechanics. We set this concern aside because at this stage in the transaction, the A+ rating is a structural load-bearing wall, and we are not paid to do demolition.

  • Residual value uncertainty. The RVG depends on modeled guaranteed minimum values that assume buyers will one day desire a vast hyperscale complex in Richland Parish under stress scenarios. If hyperscale supply balloons or the resale market for 2-gigawatt data centers becomes as illiquid as common sense, Meta will owe more money. This increases Meta’s direct obligations, which should concern us, but does not, because Meta is rated AA-/Stable and therefore presumed to withstand any scenario we have chosen not to model.

  • Casualty and force majeure. In extreme scenarios, multiple buildings could be destroyed by a hurricane, which we view as unlikely given that they almost never impact Louisiana. The logic resembles a Rube Goldberg machine built out of indemnities. We classify this as a strength.

  • JV structural subordination. Cash flows must navigate waterfalls, covenants, carve-outs, and the possibility of up to $75 million of JV-level debt. These features introduce structural complexity, which we flag, then promptly ignore, because acknowledging would force us to explain who benefits from the convolution.

Despite these risks, we maintain an A+ rating because Meta’s credit quality is strong, the structure is designed to hide risk rather than transfer it, and our role in this ecosystem is to observe these contradictions and proceed as though they were features rather than warnings.

The outlook is Superficially Stable. That means we expect the structure to hold together as long as Meta keeps paying for everything and the accounting rules remain generously uninterested in economic reality.

We assume, with the confidence of people who have clearly not been punished enough:

  • Meta will preserve an AA-/Stable profile because any other outcome would force everyone involved to admit what this actually is.

  • Construction will stay “broadly on schedule,” a phrase we use to pre-forgive whatever happens as long as Meta covers the overruns, which it must.

  • Lease payments and the minimum rent schedule will continue producing a DSCR that hovers around 1.12 in models designed to ensure that result, and not materially below 1.10 unless something un-modeled happens, which we classify as “outside scope.”

  • The RVG will remain enforceable, which matters more than the resale value of a hyperscale facility in a world where hyperscale facilities may or may not be worth anything.

  • Changes in VIE or lease-accounting guidance will affect where Meta stores the debt, not whether Meta pays it.

We could lower the rating if Meta were downgraded, if DSCR sagged below the range we pretend is acceptable, if Meta weakened its guarantees, or if events unfold in ways our assumptions did not account for, as events tend to do. The last category includes anything that would force us to revisit the assumptions we confidently made without testing.

We view an upgrade as unlikely. The structure already performs the single miracle it was designed for: keeping $27.3 billion off Meta’s balance sheet in a manner we are professionally obligated to support.

CONFIDENTIALITY AND USE: This report is intended solely for institutional investors, entities required by compliance to review documents they will not read, and any regulatory body still pretending to monitor off-balance-sheet arrangements. FSG LLC makes no representation, warranty, or faint gesture toward coherence regarding the accuracy, completeness, or legitimacy of anything contained herein. By reading this document, you irrevocably acknowledge that we did not perform due diligence in any conventional, philosophical, or legally enforceable sense. Our review consisted of rereading Meta’s press release until repetition produced acceptance, aided by a Magic 8-Ball we shook until it agreed.

LIMITATION OF RELIANCE: Any resemblance to objective analysis is coincidental and should not be relied upon by anyone with fiduciary obligations, ethical standards, a working memory, or the ability to perform basic subtraction. Forward-looking statements are based on assumptions that will not survive contact with reality, stress testing, most Tuesdays, or a modest change in interest rates. FSG LLC is not liable for losses arising from reliance on this report, misunderstanding this report, fully understanding this report, or the sinking recognition that you should have known better. Past performance is not indicative of future results, except in the specific case of rating agencies repeating the same mistakes at larger scales with increasing confidence.

RATING METHODOLOGY: The rating assigned herein may be revised, withdrawn, or denied ever existing if Meta consolidates the debt, Louisiana ceases to exist for tax purposes, or the data center becomes self-aware and moves to Montana to escape the heat. FSG LLC calculated the A+ rating using a proprietary model consisting of discounted cash flows, interpretive dance, and whatever number Meta’s CFO sounded comfortable with on a diligence call we did not in fact attend. Readers who discover material errors in this report are contractually obligated to keep them to themselves and accept that being technically correct is the least valuable form of correct.

GENERAL PROVISIONS: By continuing to read, you consent to the proposition that what Meta does not consolidate does not exist, waive your right to say “I told you so” when this unravels, and accept that the term “investment grade” is now a disposition rather than a metric. FSG LLC reserves the right to amend, retract, deny, or disown this report at any time, particularly if Congress shows interest or someone notes that $27 billion off-balance-sheet is on a balance sheet somewhere. If you print this document, you may be required under applicable securities law to recycle it, shred it, or burn it before sunrise, whichever comes first. For questions, complaints, or sneaking suspicions, please do not contact us. We are unavailable indefinitely and have disabled our voicemail.

Leave a comment

联系我们 contact @ memedata.com