The Price of Walking Away

Correction, 15 October 2026: This piece said that whether a given supplier-side guarantee is circular "depends on documents we have not read". Those documents, including AMD's and Nvidia's annual reports, Broadcom's quarterly report of 9 June and Nvidia's 8-K of 17 August, were all public before this piece was written, and should have been read first. It also said the tenant and supplier shapes "share a label and almost nothing else", which overstated the difference: where the terms are disclosed, both pay a shortfall against what the asset fetches. The details are in In Exchange for the Guarantees, dated 12 October 2026.

2 September 2026. Figures are as of the dates stated in the text.

Interest disclosure: T-OpenAI is a tokenized loan participation right providing economic exposure linked to OpenAI's pre-IPO valuation, issued through a dedicated issuing subsidiary, as are T-SpaceX and T-Kalshi. Following the SpaceX listing, the investment underlying T-SpaceX is being divested. Redemption opens only once a Redemption Start Date is announced, and as of 2 September 2026 none has been. Tessera's founder, and entities he controls, hold related exposure. This piece discusses financing arrangements involving OpenAI directly. Read what follows with that in mind.

What Is the Collateral Actually Worth?, published on 26 August, noted that SPV structures in AI infrastructure are often described as giving their sponsors the option to walk away, and that whether the description holds "depends on residual value guarantees and credit support that are, in most of these deals, not public."

In most of these deals that support is still not public. In one of the largest it is, in full, in the notes to filings anyone can read on EDGAR. Those notes repay close reading, because they show what a fully disclosed residual value guarantee actually establishes. The walk-away is not free. What it would cost is another question, and the filings cannot answer it, and will not be able to for years.

What Meta Disclosed

Meta's Form 10-K for 2025 and its Form 10-Q for the quarter ended 30 June 2026 describe the company's data center venture in Louisiana in unusual detail. Meta holds a 20% membership interest, and the parties have committed to fund their pro rata share of approximately $27 billion of total estimated development costs. Meta's leases commence in 2029 with an aggregate initial lease commitment of approximately $12.31 billion; each property carries an initial four-year term, with renewal options running to a total of up to twenty years.

The notes then answer the question the 26 August piece raised. Meta has provided residual value guarantees (RVGs) with an aggregate threshold of approximately $28 billion, a threshold that decreases over time. The trigger is stated in the company's own words: "If we decide to terminate or not renew a lease, and if certain other conditions are met, our maximum RVG payment would equal any shortfall between the fair value at that time and the RVG threshold for that property." Meta adds that such payments are not probable, and that no liability has therefore been recorded. The same filing describes the venture as providing "strategic optionality and flexibility."

One paragraph later comes a number. Because Meta does not consolidate the venture (it is not the primary beneficiary), it must disclose its maximum exposure to loss. As of 30 June 2026 that figure was $46.03 billion, up from $45.95 billion at 31 December 2025, "consisting of the carrying value of our equity investment, the lease commitments, our estimated future funding commitments, and the maximum RVG threshold."

The Trigger Prices the Exit

Read the trigger carefully. The guarantee applies only if Meta decides to terminate or not renew. It is not a general mark to market on the asset. It is the price of exercising the option to leave, and nothing else.

That settles the question the 26 August piece left open, in the narrowest possible way. That piece argued that where the sponsor stands behind the paper there is no true walk-away, and the ring-fence is presentational. On this deal the sponsor does stand behind the paper, in writing, up to a stated ceiling. The option to leave exists, and it has a price. Both facts sit in the same sentence of the same filing, and most descriptions of these structures repeat the first without the second.

The "no liability has been recorded" line deserves care, because it is being handled loosely elsewhere and is not evidence of anything untoward. A contingency that is not probable is not recorded. That is the ordinary accounting test, applied normally, with the arrangement disclosed in the notes where such things belong. Presenting the absent liability as concealment overreaches; the disclosure here is not the problem, and is in fact unusually complete. The problem is subtler, and the earlier piece did not anticipate it. A ceiling is not a price.

What the $46.03 Billion Figure Measures

Consider what the figure is made of: the carrying value of the equity investment, plus the lease commitments, plus the estimated future funding commitments, plus the maximum RVG threshold. It is an arithmetic sum of components, each taken at the worst assumption available, with the guarantee paying out at its full, undeclined ceiling on top of every lease payment and every funding call. That is what a maximum-exposure disclosure is for, and it is the right disclosure. It marks the top of the range, and Meta does not present it as an estimate of what will happen.

What the filing does not provide, and what no filing could, is a way to locate a point inside that range. Three variables stand between zero and $46.03 billion, and all three are unresolved.

The first is when. The leases commence in 2029 with four-year initial terms, so the earliest point at which a termination or non-renewal could trigger a payment sits in the 2030s.

The second is how far the threshold has fallen by then. It decreases over time on a schedule that is not published. The $28 billion is the aggregate at the start; the figure that matters is whatever remains on the day a decision is made, and outside the company that figure is unknown.

The third is what the asset is worth at that moment. The exposure is a shortfall against fair value, which means the fair value, in the 2030s, of purpose-built single-tenant AI infrastructure. The 26 August piece argued that lenders had sidestepped the hardware's economic life by underwriting the customer instead, so that nobody had been required to hold a view on it. A campus poses the same problem over a longer horizon, with no offtaker in between.

Meta, then, has published a number, prominently, and to a standard the rest of this sector does not meet. But the number is a bound, the distribution beneath it is empty, and a bound is what a company discloses when the distribution is the thing it cannot compute.

How This Differs From the CoreWeave Facility

The 26 August piece described the $2.6 billion facility CoreWeave closed on 10 August as the first structure we have seen in this market where somebody underwrites residual value on the equipment itself, and noted that the question had gone largely unasked because the offtaker structure made asking it unnecessary.

Meta's guarantee is the same question at the campus level, and it was asked and answered months earlier; the venture dates from October 2025. But the two are not the same instrument, and they should not be added together. A lender taking residual risk on GPUs has priced something, however roughly. A sponsor writing a residual value guarantee has agreed to pay a difference that nobody has priced at all, on a date that may fall beyond the useful life of the hardware the campus was built to house. The first is a market forming; the second is a market being deferred.

Two Structures Under One Label

A distinction is getting lost in the coverage of these arrangements. Meta guaranteeing the residual on a campus it will occupy is a tenant pricing its own exit: the guarantee is what it costs Meta to stop using something Meta uses.

A structurally different arrangement is filed under the same heading. When a supplier guarantees the residual on a facility built to house that supplier's own product, the guarantee underwrites demand for what the supplier sells. That is the structure The Money Goes Round, published on 28 August, examined in a different instrument. The two shapes share a label and almost nothing else.

Meta's arrangement has sometimes been described as circular financing; on the filing, it plainly is not. Whether any given supplier-side arrangement qualifies depends on documents we have not read, and this piece does not characterize any specific one on the strength of reporting about it. The distinction is the first question to ask of any of these structures, and it is routinely skipped.

What This Does and Does Not Establish

This piece predicts no call on any guarantee. It alleges no inadequacy of disclosure; the disclosure here exceeds the sector's norm. It makes no claim that the assets are bad.

What it establishes is that the caveat in the 26 August piece can be closed for at least one deal, and that closing it resolves less than might be expected. Where the guarantee is public, it confirms that the ring-fence is presentational, and Meta goes further than the caveat required by publishing a maximum exposure of $46.03 billion. What that gives a reader is a ceiling. What it cannot give is any sense of where inside the ceiling the answer lies, because the schedule that would narrow it is not published, and the market that would price it does not exist.

The 26 August piece argued that when these structures go wrong the loss lands on the credit layer, and that the sponsors' cushion is thinner than the headline balance sheets suggest. One line is worth adding to that now. The sponsors also carry a dated, declining, unquantified obligation that crystallizes only if they choose to leave, which is the cheapest kind of obligation to carry and the hardest kind to see.

Sources: the Tessera pieces of 26 and 28 August 2026, What Is the Collateral Actually Worth? and The Money Goes Round, as published. All figures for the Louisiana venture, including the 20% membership interest, the approximately $27 billion of total estimated development costs, the leases commencing in 2029 with an aggregate initial lease commitment of approximately $12.31 billion and initial four-year terms renewable to a total of up to twenty years, the residual value guarantees with an aggregate threshold of approximately $28 billion decreasing over time, the termination-or-non-renewal trigger, the statements that RVG payments are not probable and that no liability has been recorded, the description of the venture as providing strategic optionality and flexibility, and the maximum exposure to loss of $46.03 billion as of 30 June 2026 and $45.95 billion as of 31 December 2025 with its four stated components, are from Meta Platforms Form 10-K for the year ended 31 December 2025, filed 29 January 2026, and Form 10-Q for the quarter ended 30 June 2026, filed 30 July 2026, both read from EDGAR on 30 August 2026. The CoreWeave facility and telecom-era vendor financing are as cited in the 26 and 28 August pieces. Figures are as of the dates given.

This is market commentary, not investment advice. It is not a recommendation to acquire, hold or redeem any Tessera product, or to take or avoid exposure to any company mentioned.

T-Tokens are tokenized loan participation rights, not equity. High risk. DYOR. Not financial advice. Not available in the US or other restricted territories. tessera.pe/terms