Meta’s AI Building Boom Has a Risk Wall Street Cannot Keep Off the Books

By Nkozi Knight | October 25, 2026

Americans should not have to understand a maze of limited liability companies to know whether their retirement savings are helping finance Silicon Valley’s next enormous bet. Yet Meta’s artificial intelligence expansion raises exactly that concern. In Richland Parish, Louisiana, a massive data center project is becoming a case study in how construction costs, corporate commitments and investment risk can be spread among different entities, leaving the public with a much simpler picture than the economics deserve.

The question is bigger than whether Mark Zuckerberg can build the computing infrastructure he wants. It is whether Wall Street is financing that ambition in ways that could leave ordinary investors carrying risks they never realized they had accepted.

Construction began on a 2,250-acre site. Meta initially announced more than $10 billion of investment in December 2024. Its current Louisiana project page advertises more than $50 billion and five gigawatts of computing capacity. That dramatic increase deserves scrutiny, but precision matters: the project has expanded, so these numbers do not establish that an unchanged construction budget simply ran five times over its original estimate. [1][2][3]

In October 2025, Meta announced a venture in which funds managed by Blue Owl Capital would own 80% and Meta would retain 20%. The announcement identified approximately $27 billion in development costs for buildings and long-lived power, cooling and connectivity infrastructure. That figure covers a particular financing scope, rather than every dollar associated with the broader AI campus. Meta contributed land and construction assets; outside capital would help finance development, including debt purchased by PIMCO and other investors. Meta would lease the facilities. The construction therefore is not funded entirely from Meta’s corporate cash, but neither is Meta contributing nothing. [4]

Beignet Investor LLC is the name retirement investors should learn. S&P Global Ratings assigned its $27.3 billion senior secured debt an A+ rating with a stable outlook in October 2025. The rating was not A-minus, and EY did not issue it. EY audits Meta. S&P rated this debt. Moody’s has separately examined the accounting and credit risks surrounding these arrangements. Those distinctions matter because an audit opinion, a bond rating and an assessment of systemic risk answer different questions. [5]

Page 106 is worth reading. So is what comes after it.

In Note 5 on printed pages 106–107 of Meta’s 2025 annual report, Meta disclosed approximately $12.31 billion in initial lease commitments, a roughly $28 billion declining residual-value guarantee threshold, and maximum venture exposure of $45.95 billion. The company recorded its equity investment as an asset and did not consolidate the venture, saying it lacked power over the activities most important to its economic performance. It recorded no guarantee liability because payment was not considered probable. This is not debt transformed into an asset; it is an equity investment recorded while the separate entity’s borrowing remains outside Meta’s consolidated balance sheet. [6]

EY identified the consolidation decision as a critical audit matter, explaining that it required especially challenging judgment. That merits attention, but it was not an adverse audit opinion or a finding of fraud. EY still concluded that Meta’s financial statements fairly presented its finances under U.S. accounting principles. [6]

The newer disclosure is even more revealing. Meta’s September 2026 filing describes both Louisiana and Texas ventures, each 20% owned by Meta. Together, they carried $68.4 billion in disclosed maximum exposure, including approximately $41.5 billion in residual-value guarantees. The company still reported no liability for those guarantees because payments were not considered probable. These are maximum exposure figures, not expected losses or amounts immediately payable. [7]

The real concern is the distance between accounting presentation and economic responsibility. A company can comply with reporting rules while investors still need substantial additional analysis to understand what it might ultimately have to pay.

Moody’s has identified that problem. Its February 2026 warning, reported by the Financial Times, explained that accounting can leave a gap between potential lease-renewal costs and potential payments for not renewing. Moody’s said it would make its own assessments and adjust debt where reported lease liabilities understated likely cash outflows. Separately, Moody’s Analytics highlighted how renewal assumptions and residual-value guarantees can leave economic exposure incompletely reflected in accounting metrics. [8][9]

In plain language, the economics may still involve expensive choices even when the balance sheet does not yet show the full cost. Continuing to use a facility costs money. Leaving it may also cost money if contractual guarantee conditions are met. Moving borrowing into another legal entity does not make those choices disappear.

That is the accounting gap policymakers should address. The case for reform does not require accusing Meta of breaking existing law. It requires asking whether existing rules make substantial obligations clear enough to the people whose money is at risk.

The retirement connection is real, but it must be described honestly.

A PIMCO Income Fund fact sheet distributed through GuideStone lists Beignet among its largest holdings, at approximately 1.81% of net assets as of June 30, 2026. BlackRock’s Multi-Asset Income Portfolio also identifies Beignet debt in its January 2026 shareholder report. These are documented examples of the bonds reaching investment portfolios beyond the immediate project sponsors. [10][11]

A worker whose retirement account owns an exposed fund can therefore have an indirect stake in this financing without ever choosing a security called Beignet. Pension exposure likewise depends on the actual funds and mandates a plan owns. It would be inaccurate to claim that every 401(k), every pension or every BlackRock fund holds these bonds.

Position size matters. Hypothetically, if a fund had 2% invested in a bond that lost half its value, the direct impact would be about 1% of the fund before other changes. One position does not automatically erase a retirement account. The more serious possibility is that investors own several apparently different funds whose stocks, bonds and private investments depend on the same AI spending cycle.

That is how diversification can disappoint: different fund names can still conceal overlapping economic bets.

The $3 trillion figure is alarming enough without mislabeling it.

The Wall Street Journal reported roughly $3 trillion in off-balance-sheet commitments across nine technology companies, including future leases and purchase obligations for infrastructure, chips and other needs. That does not mean Meta, PIMCO and BlackRock jointly hold $3 trillion of hidden debt, or that $3 trillion has already been packaged into A-rated bonds. Borrowing, future purchases, guarantees and leases are related but different obligations. Adding them indiscriminately risks double counting and exaggeration. [12]

The warning is that future cash demands can be much larger than headline debt figures suggest. A construction loan and a tenant’s commitment to pay rent may be two sides of the same financing arrangement. Regulators and investors need to trace who ultimately bears each risk, rather than simply adding every impressive number together.

Is this the next Enron? The comparison is a warning, not a verdict.

Enron’s history makes the question understandable. Its use of special-purpose entities became part of a documented fraud scandal involving concealed relationships and misleading financial reporting. Former chief financial officer Andrew Fastow pleaded guilty to conspiracy charges. Meta’s disclosed venture structure is not, by itself, evidence of comparable wrongdoing by Zuckerberg or anyone else. [13]

The useful lesson is about scrutiny. Complex structures deserve more attention when a company receives substantial economic benefits while major financing obligations sit elsewhere. The existence of an SPV does not establish fraud, but legal separation should never become an excuse to stop following the money.

The mortgage-crisis comparison also has limits. These bonds are not a pool of subprime home loans. The relevant concern is whether investors rely too heavily on ratings and contractual protections while underestimating a shared downside scenario. An investment-grade rating is an assessment of credit risk, not a guarantee against price declines, technological change or losses.

There are meaningful protections here. Meta’s lease commitments and guarantees support the financing, and the company has a substantial operating business. In its latest quarterly results, Meta reported $64.6 billion in revenue, up 26% from a year earlier. That financial strength matters when evaluating the difference between a risky investment cycle and an imminent corporate collapse. [4][14]

But the risks are not imaginary. Moody’s has pointed to long investment payback periods and technological change. Its later research also discusses construction, power delivery and operational delays that can upset financing assumptions. A completed building is not automatically a profitable AI business, and valuable infrastructure today may require substantial additional investment to stay competitive tomorrow. [9][15]

Here is how trouble could spread.

If AI revenue disappoints, companies could cut expansion plans and become more selective about renewing capacity. Lower demand could weaken project values. Bond prices could fall, refinancing could become more expensive, and investors facing withdrawals could sell other assets. Guarantees might bring obligations back to corporate cash flows just as operating conditions deteriorate. This is a plausible stress scenario, not a prediction that every step will occur.

On October 19, Reuters reported S&P’s warning that interconnected AI financing may create more liabilities than are currently visible. S&P also said it did not then see a reason for ratings to change. Both points belong in this discussion: visibility is imperfect, but a recognized risk is not the same as an agency forecasting collapse. [16]

By 2028, investors may have much more evidence about whether today’s investment promises are producing adequate returns. But a claim that Americans’ life savings will be wiped out by that date would go beyond the evidence. Trump’s term is scheduled to end in January 2029, and political calendars do not determine credit cycles. Meta’s Louisiana leases are scheduled to begin in 2029; its Texas leases begin in 2028. Stress could emerge earlier through prices and financing conditions, or later through operating results and renewal decisions. [7]

What should change now is the standard of transparency. Companies should provide a clear reconciliation between reported debt and their major financing-related commitments, including scenarios for lease renewals and guarantees. Fund managers should explain concentrations across the AI ecosystem, including exposure through unfamiliar issuers. Retirement-plan fiduciaries should assess how those holdings behave together under stress.

For individual investors, the useful response is to inspect holdings, ask about exposure and evaluate concentration, rather than make panic trades based on a dramatic headline. A fund’s name is not a complete explanation of what it owns.

Zuckerberg may help build enormously productive infrastructure. He may also help establish financing practices that spread the consequences of an overinvestment cycle far beyond Silicon Valley. We do not yet know which outcome will dominate.

We do know that keeping debt outside a company’s consolidated balance sheet does not keep the economic risk outside the financial system.

Americans deserve to understand where that risk lands before the losses arrive in our retirement accounts.

Sources

1. DPR, Richland Parish construction site

2. Meta, original December 2024 announcement

3. Meta, current Louisiana campus investment and capacity

4. Meta, October 21, 2025 joint-venture announcement

5. S&P, October 23, 2025 A+ debt rating

6. Meta 2025 Form 10-K, printed pp. 83, 86, 106–107

7. Meta September 2026 Form 10-Q, Note 5, printed pp. 17–18

8. Financial Times, Moody’s accounting warning, February 23, 2026

9. Moody’s Analytics, Power without delivery, July 31, 2026

10. GuideStone, PIMCO Income Fund fact sheet, holdings June 30, 2026

11. BlackRock Multi-Asset Income Portfolio, January 31, 2026 shareholder report

12. Wall Street Journal, Why Big Tech’s AI Spending Is $3 Trillion Higher Than It Seems

13. SEC, Fastow guilty plea and civil fraud settlement, January 14, 2004

14. Meta, third-quarter 2026 earnings

15. Moody’s Ratings, October 27, 2025 credit outlook

16. Reuters, S&P warning, October 19, 2026

Leave a Reply