You underwrote the tenant's credit and the tenant's power. You did not underwrite the tenant's lender.

Current as of September 14, 2026. The PUCT Project 58481 rule discussed below is a Staff Recommended Adoption Order, not an adopted rule. Confirm the final text before acting on it. Revision log at the foot of this article.

Part of a five-article series on phased Texas data center campus development and the commitments sized to a buildout that may not arrive.

• Part one. The exposure: why a compliant tenant is the expensive one
• Part two. Where the exposure bites: security, tariffs, hedges and covenants
• Part three. Sizing the shared works against an uncertain phase 2
• Part four. Public commitments sized to full buildout
• Part five. Recovery once the phase has already stalled

The short version, for the capital committee.

Everything else below is negotiation.

---

The failure mode is rarely a default. It is a tenant that keeps paying and stops expanding.

A defaulting tenant gives you remedies, a claim, security and a reletting process. A tenant that performs on phase 1 and quietly declines phase 2 gives you none of those, and leaves you worse off, because the shared works you sized for the full campus stay on your balance sheet while the counterparty stays fully compliant. There is no breach to declare. Nothing triggers. Most campus documents are drafted heavily for the first case and lightly for the second.

Standard tenant diligence asks whether the counterparty is good for the rent and whether it has a credible path to power. Both are answerable. Neither answers the question that decides whether your oversized substation position ever gets a funded beneficiary: how is this tenant paying for its own buildout, and what has to stay true for that funding to continue. The answer is not in your lease file, and over the last two years it has moved somewhere your diligence does not reach.

The useful part is that much of the protection available here needs no counterparty agreement at all. You will not win a disclosure fight against a hyperscaler, so the sections below do not ask you to. What follows separates what you can impose on your own capital committee from what you have to negotiate, and what to do when they decline the negotiated part, which they often will.

What the financing evidence actually says

The Bank for International Settlements published Bulletin No 120, "Financing the AI boom: from cash flows to debt," by Iñaki Aldasoro, Sebastian Doerr and Daniel Rees, on 7 January 2026. Its key takeaways, quoted rather than paraphrased:

AI-related investment is surging both nominally and as a share of GDP, and currently accounts for a substantial share of economic growth. The size of anticipated investment needs will require firms to shift the source of financing from operating cash flows to debt, with private credit playing a rapidly increasing role. Macroeconomic and financial stability risks appear moderate, but the boom's sustainability hinges on AI firms meeting high earnings expectations. Equity prices have run far ahead of debt market pricing.

Risk is moderate, not acute. This is not a distress warning, it is a change in the source of financing, and the source determines how a counterparty's capacity to spend behaves when conditions move.

One number carries most of the weight for a landlord. In building its estimates the Bulletin adopts a figure from Noffsinger and co-authors, that data center equipment investment runs at roughly three times data center construction investment. Your tenant's invisible spend, the servers and the fit-out, is several times the visible building you are leasing them. That ratio is why the tenant's expansion decision is far more sensitive to credit conditions than their rent obligation is. Rent on a running hall is serviced from revenue. The next 60 MW of compute is serviced from a facility that has to be available and drawable at the moment the expansion option comes due. Those are different exposures, and only the first is underwritten in your file.

The BIS Quarterly Review of 14 September 2026 is a useful later marker: it described unease about valuations and possible overinvestment in the technology sector disrupting AI-driven equity momentum, while noting risk appetite proved resilient on the whole.

The mechanism, and why your diligence file cannot see it

When a tenant funded expansion from operating cash flow, counterparty analysis was reasonably complete. The entity signing the lease generated the cash, and the parent financials described the capacity to spend. Debt-funded expansion breaks that alignment. Capital is raised into structures beside the operating parent rather than inside it: special purpose entities holding a single campus or tranche of equipment, joint ventures with an infrastructure or financial sponsor, private credit facilities with covenant and drawdown mechanics, and operating leases that keep equipment commitments off the visible balance sheet of the entity you are looking at. These are ordinary financing choices. The landlord's problem is narrower: they are not visible in the documents tenant diligence conventionally collects.

So you can hold a signed long-term offtake from a creditworthy counterparty whose ability to perform on phases 2 through 4 depends on a financing chain you have never examined, cannot see in the signing entity's audited statements, and have no contractual right to be told about. Capacity to expand is not capacity to pay rent, and only the second is in your file.

The behavioral consequence, which is inference rather than a finding: when expansion is funded by drawn facilities rather than accumulated cash, its pace becomes sensitive to credit conditions in a way that rent on a running hall is not. Nothing about that predicts distress at any particular firm. It changes which of your tenant's obligations are robust and which are elastic, and the elastic one is the one your shared works are sized for.

Be honest about where this bites hardest, because it is not uniform. A top-tier hyperscaler funding expansion at a rated parent has visible debt, and your exposure there is real but bounded by disclosure you can actually read. The concern sharpens considerably for neoclouds, sponsor-backed developers and joint-venture vehicles, where the expansion capital sits at an entity with no public reporting obligation and the signing entity may be a matter of weeks old. A campus anchored by a rated hyperscaler and a campus anchored by a sponsor-backed operator are different underwriting problems wearing similar lease documents.

Where standard campus underwriting misses it

The standard discipline: a phase can be financeable as a building while the main, road or electrical work enabling it remains unfunded. So you run delayed occupancy and no further expansion through the same funding case, then establish who carries unused capacity, operations and any unrecovered advance.

It is incomplete in one respect. It treats "later phases may not arrive" as a risk to stress-test tenant by tenant. It does not ask whether the probability of arrival is correlated across your tenants.

Demand-side correlation is already standard in most funding cases. Correlation through a shared credit channel generally is not, and the financing shift is precisely that channel. If phases 2 through 4 are anchored to counterparties funding expansion from drawn facilities priced against the same set of AI earnings expectations, their delays are not independent draws. One condition moves several tenants at once, in the same direction, at roughly the same time.

You can test the magnitude this week without new data, using the no-further-expansion case you already run. It almost certainly assumes tenant slippage is independent. Re-run it with a shared factor driving the tenants funded through debt and see how far the simultaneous-slip scenario moves. It will not be a small move, and that second run is the one your lender is underwriting whether or not you have produced it. The full specification, tight enough to hand to an analyst without a follow-up meeting, is in the appendix below.

A practical screen: sort your anchor tenants by how they fund expansion, not by credit rating. Counterparties funding from operating cash flow, counterparties funding through corporate debt, and counterparties funding through project-level or sponsor vehicles are three different exposures. A campus anchored entirely in the third category is a concentrated position, no matter how strong each individual name looks.

What happens, in sequence

Trace the non-expansion case through a phased campus and it compounds in a specific order. Shared works here means every commitment sized to the full buildout: the substation and electrical backbone, yes, but also the water and wastewater supply contract and any reserved capacity in it, the Chapter 380 or 381 development agreement and any associated abatement, the carrier IRU, the land takedown schedule and the EPC guaranteed maximum price scope. The electrical position gets examined hardest. That is precisely why it is the least likely place to find an unmanaged exposure. The mechanisms below apply in order.

Oversized shared assets with no funded beneficiary. The substation position, main diameter and road section were specified for full-campus load. Phase 1 uses a fraction of the capacity and carries all of it. If phase 2 never funds, you hold a permanent asset sized for revenue with no committed source. No breach. Nothing triggers. The gap just sits there.

The instruments that carry it are four, and you have signed all of them. Posted security under ERCOT's Batch Zero process, roughly $15 million on a 300 MW position. A non-utilization clause in your own TSP tariff that can recalculate CIAC on realized load. A construction hedge sized to a drawdown that may not happen. And a covenant test where your lender asks whether phase 1 revenue services the oversized asset.

Each has its own clock, trigger and counterparty, and they are covered in Four instruments decide what a stalled phase 2 costs you. Two findings from it bear on the arithmetic below.

The regulatory clock moved from six months to twenty-four, and the adopted rule returns 80 percent of associated security rather than forfeiting it, though the retention is tiered and Batch Zero projects failing the maturity test lose 50 percent. A twenty-four month runway means the regulator will not surface a stalled phase on any timeline useful to you.

And your own tariff may already contain the clawback. Oncor's non-utilization clause recalculates CIAC against realized load within four years of extension completion, and the charge lands on you rather than the tenant.

The shared works are not just the substation, and the rest of them are worse. Apply the same phase 2 question to the water contract, the development agreement and the district assessment, which are covered in Your electrical position gets examined. Impact fees are not refundable because your own phases did not arrive. Reserved water capacity is not the standing charge most people assume, since none of the three executed Texas agreements reviewed carries take-or-pay. Incentive clawbacks range from full recapture to expressly none.

The clawback and the carry model are the same exposure, which is the part that gets missed. The carry table below includes a property tax line at 2.2 percent, and the reflexive objection is that an abatement removes it. Look at when the abatement is conditioned. If eligibility is tied to an investment or jobs number sized to the full buildout, the non-expansion case is the case where it lapses. The tax line does not just persist, it can arrive retroactively as recapture at the same moment the carry is running and the position is vacant. On an $84 million unrecovered position, losing a full abatement moves carry from roughly 9.8 percent to 12 percent, with recapture of prior years landing as a separate claim on top.

That is a compounding failure, not two independent ones. The tenant's decision not to expand triggers your shortfall against the public agreement, which raises the carry on the asset you are already holding vacant, in the period when your coverage ratio is worst.

Default versus non-expansion, side by side. The two cases are not variations of the same event. They differ on every dimension that matters to a recovery.

The second column is the one that costs more, and it is the one most campus documents barely address.

Now the arithmetic, and it has to be a present value. Comparing an annual carry figure to a one-time damages claim is not a comparison, and an earlier version of this article made that mistake. A recurring cost and a single recovery are different objects. Put both in present value, net the residual, and the picture is worse than the ratio suggested, not better.

Work a hypothetical campus: 300 MW planned, phase 1 at 60 MW, shared works built for the full 300 MW at $350,000 per MW. The 240 MW of unrecovered capacity is about $84 million of substation position, main, road and backbone.

First, define carry properly, because 9 percent is the cost of capital and not the cost of holding the asset.

Carrying an oversized position at the cost of capital alone understates the annual number by about $2.5 million, a third of the way low. The property tax line deserves particular attention, because the obvious objection is that an abatement removes it. Sometimes. But a Chapter 312 or 380 abatement is commonly conditioned on the investment or jobs number tied to the buildout, which is precisely what is not arriving in this scenario. The non-expansion case is the case where the abatement is most likely to lapse or be recaptured. Run the tax line both ways and know which side of it you are on before you assume it away.

Now the loss, as a present value, discounted at the 9 percent cost of capital. The position is not worthless. In a tight ERCOT market an energized 240 MW position relets, so the honest exposure is the timing loss: carry borne during the vacancy, plus the revenue pushed out by it.

Against that, the remedy. Delay damages on phase 2 are typically liquidated and capped at a few months of the unbuilt phase's rent. At $110 per kW-year on a 60 MW phase 2, a six-month cap is about $3.3 million. At $150 it is $4.5 million.

So at an eighteen-month vacancy the remedy covers about 7 percent of the loss. At three years it covers about 4 percent. The ratio is not 2.3 times. Properly computed it is thirteen to twenty-five times, and the earlier figure was low because it compared one year of carry to the whole claim while ignoring deferred revenue entirely.

Two things that table does not include, both of which should be run separately rather than buried in the headline. If the relet happens at a permanent discount rather than at the same rent, add the present value of that haircut: a 15 percent haircut over twenty years is about $37 million on top. And if the position does not relet at all, the loss approaches the full capital plus carry, which is the scenario that ends campuses rather than bruising them.

The correlation question and the relet question are the same question. Every row in that table is driven by the vacancy duration, and vacancy duration is driven by how many plausible takers are funded the same way as the tenant who just declined. That is why funding-type concentration in your leasing pipeline drives this loss more than any lease term does.

One more thing about the $350,000 per MW. It is partly your decision, not purely an input, because you choose how much to oversize. Every figure here is computed at full-campus sizing, the most exposed end of the range. Phase-gating moves you down that axis. So does designing for expansion, at a fraction of the cost, which the later section prices.

Every figure above is hypothetical, chosen to show shape rather than to price a campus. The input to substitute first is vacancy duration, because the table is far more sensitive to it than to the carry rate: moving carry from 9 percent to 12 percent changes the three-year loss by about 6 percent, while moving the vacancy from eighteen months to three years nearly doubles it. Most underwriting spends its energy on the wrong one of those two.

What you can require without anyone's consent

Phase-gate infrastructure sizing to the funding, not the plan. Separate minimum opening-phase scope from growth capacity, then attach the growth increment to a verifiable trigger: an exercised expansion option with a deposit, a funded CIAC contribution, or a take-or-pay on capacity rather than space. Oversizing is still frequently correct, and the next section makes that case properly. Oversizing with no named trigger is the specific practice to stop.

Answer on paper who funds the delta if phase 2 never arrives. One number, one named party. If your capital committee cannot produce both before the growth increment is authorized, it is not ready for approval. It is also the cheapest thing here to put in place.

Re-run the no-expansion case with correlated slippage. Your current case almost certainly assumes independence. The specification is in the appendix.

Use your own lender as the negotiating position. "I would like to understand your capital structure" is a preference a counterparty can decline. "My construction lender requires a named funding source for the growth increment before it will fund shared works" is a condition, and it is usually true. Have the conversation with your lender first and let their covenant requirements set the frame. It converts an awkward request into a constraint neither side controls. Use it once and hold it. Concede the point after invoking your lender and you have taught the counterparty that your lender is decorative.

Hold the evidence. The entity chart showing which party would hold expansion capital. The expansion trigger and its verification mechanism. The capacity-release provision mapped to the applicable utility milestone. The no-further-expansion case run with correlated slippage. A written record of who carries unused capacity, operations and any unrecovered advance.

The case against everything above

Everything to this point is written from the balance sheet. The development argument runs the other way, and it is not weak. It is made properly in a companion article, Most campus budgets have two categories for shared works. The money is in the third, which prices the retrofit premium, the forfeited queue slot and the middle path this article's framing leaves out.

Three findings from it bear directly on the argument here.

Retrofit is not the same cost. Substation equipment removal runs about what installation cost, and duct bank retrofit buys the excavation twice at roughly 13 times material cost. On a 240 MW increment the build-twice premium commonly lands between $8 million and $15 million, which is one to two years of the carry computed above. Phase-gating pays off only if the slip runs past that window.

The queue slot is a separate loss. It took longest to obtain, cannot be re-entered mid-cycle, and in a tight ERCOT market its economic value routinely exceeds the retrofit cost. A developer who phase-gates too aggressively can protect the balance sheet and lose the campus.

And there is a third column. Designing for expansion, provisioning the route without installing the capacity, costs roughly $2,000 to $6,000 per MW against $10.1 million a year of full carry or $8 to $15 million of retrofit. That is an order of magnitude cheaper than either extreme, and it is the column most budgets do not have.

So the honest position is narrower than this article's recommendation taken alone. Oversizing is often correct. The practice to stop is oversizing with no named trigger, no priced option and no written answer to who carries the delta.

What to negotiate, and what to do when they decline

Price the option the tenant holds. If phases 2 through 4 are optional, charge for the option. A capacity reservation fee, a standby charge on reserved megawatts, or a declining-balance CIAC credit each convert a free option into a priced one and produce cash while the beneficiary decides. This is the ask most likely to succeed, because it trades cleanly against headline rent. Expect to concede basis points on rent to win it, and treat that trade as the negotiation rather than a concession.

Build capacity-release mechanics before you need them. This is the most valuable term in this article and it deserves more than a sentence, so it gets its own section below.

Then assume they decline the rest. A hyperscaler anchor makes the campus, and walking away over a guarantee is rarely the right call. The protection then has to come from structure rather than contract.

Size shared works to a load you could relet rather than to the tenant's stated plan, and treat the difference between those two numbers as your actual speculative position. Stage utility commitments and CIAC so exposure accrues against milestones the tenant has funded rather than milestones it has forecast.

Then deal with funding-type concentration honestly. You do not select your phase 1 anchor's funding structure. On a 300 MW campus you take the counterparty that clears your lender's coverage test and arrives with a signed LOI, and in this market that is a short list, most of it funded the same way. Treating funding type as an anchor selection criterion is advice for a market that does not exist.

Where it becomes executable is phases 2 through 4, and as a priced constraint rather than a preference. Record your anchor's funding channel at signing. Carry it into the leasing plan for later phases as a concentration limit, the same way you would treat a single-industry tenant mix. Disclose it to your construction lender before they ask, because a lender who learns that all four phases depend on the same credit channel during a covenant test will price it far more harshly than one who was told at underwriting. Where you have any choice on phases 3 and 4, a colocation provider, an enterprise or a sovereign buyer whose capital is not priced off AI earnings expectations does more for your downside than a covenant you were going to lose in negotiation anyway.

This also reframes the lapse itself. In a tight interconnection market a released 240 MW position is not purely a loss, and reletting it to a different counterparty type can beat the original tenant exercising, because it breaks exactly the concentration described above. That is only true if you planned for it. Unplanned, the same event is the three-year vacancy row in the table above, and roughly $83 million.

If it already happened

If your phase 2 has already gone quiet, the recovery options are covered in You cannot sell the capacity. You can sell the project that holds it. The short version matters to the sizing decision above, so it belongs here too.

Capacity reassignment is prohibited. The adopted Section 25.194 bars both DSPs and TSPs from reallocating transmission capacity, and only ERCOT reallocates, in a future study. You get a partial return of financial security through an administrative process, no right to nominate a successor, and no share of the capacity's value.

What does transfer is the interconnection request itself, at the entity level, without losing Batch Zero classification, provided the new entity meets eligibility criteria. So the position moves when the project moves. Recovery is a land-and-entity sale or a relet, priced against a recent Texas benchmark near $467,000 per MW, not a regulatory fix.

That is why vacancy duration drives the table above rather than any lease term. There is no fast exit from a stranded position, only a slower or quicker relet.

The next action

Not an analysis. A markup exercise your development lead can run this week.

Pull the single line and the civil drawings. Mark every asset sized above phase 1 requirement. Not the intent, the installed or committed capacity: conductor and transformer rating, duct bank ways, pipe diameter, road section, pad and foundation provision, easement width.

List every third-party commitment sized to the full campus, not just the electrical ones. The TSP agreement. The water and wastewater supply contract and any reserved capacity in it. The Chapter 380 or 381 development agreement and any abatement. The carrier IRU. The land takedown schedule. The EPC guaranteed maximum price scope. Each of these was negotiated against a buildout number, and each carries its own consequence if that number does not arrive.

For each line, record three fields. The release term, if one exists. The clawback or forfeiture on non-performance. And the date it becomes irreversible, meaning the point after which cancelling costs more than continuing.

Two lines will not be where you expect them. Your TSP's facilities extension agreement may contain a non-utilization clause with its own multi-year clock, and any impact fee already paid is almost certainly gone under Chapter 395 regardless of whether your phases arrive. Pull both before the meeting rather than during it.

Then compute carry. Do it last, deliberately. Carry is usually the smallest number on that page and always the easiest to fix, because it is yours alone and needs no counterparty. The clawbacks, the reserved-capacity charges and the irreversibility dates are the exposures that require someone else's agreement to change, which is why they need the longer lead time.

The page that comes back will not be balanced. Most campuses find the electrical position was examined carefully and the water contract, the development agreement and the takedown schedule were signed against the same optimistic buildout with nobody asking the phase 2 question at all.

---

Appendix: the correlated slippage specification

For the analyst rather than the committee. This is the run described above, specified so it can be executed and defended without further instruction.

Keep the per-tenant slip probability already in your no-further-expansion case as the marginal. Run pairwise correlation among debt-funded anchors at 0.3, 0.5 and 0.7, and report the probability of three or more anchors slipping inside the same eighteen-month window. Hold the marginal fixed across the three runs and vary only the correlation. State N explicitly, since three of four anchors is a different claim than three of eight. Name the dependence structure, a one-factor Gaussian latent model being the usual choice.

Do not load every debt-funded anchor on the factor equally. This is the instruction most likely to be missed and it changes the answer. A rated hyperscaler drawing on a revolver and a sponsor-backed operator on a private credit facility are not exposed to the same tightening in the same degree. Either restrict the correlated block to the non-rated anchors, or run two loadings, a lower one for rated parents and a higher one for sponsor and project vehicles. Pairwise correlation alone does not pin down the joint tail, and your lender's analyst will pick a structure if you do not.

Two caveats to carry into the room with the output. There is no observable dataset of correlated campus phase slippage, so the ladder is a sensitivity and the finding is the shape of the curve, not a probability. Present it that way rather than defending a point estimate. And a Gaussian latent model is thin-tailed, which understates precisely the clustered outcome this exercise is testing for, so treat the result as a floor.

On the parameters: eighteen months is roughly the period over which a credit tightening transmits to capital plans, and three or more is where a campus funding case usually breaks rather than bends. Independent draws push that scenario toward negligible. Correlation in that range moves it into the range you have to size for.

---

Revision log

September 14, 2026, revision 7. Split again into a five-article series. The instrument mechanics, posted security, the TSP non-utilization clause, the construction hedge and the covenant test, moved to their own article along with the counterparty diagnostic and the capacity release provision, leaving this one on why the exposure exists and what it costs.

September 14, 2026, revision 6. Split into a four-article series. The municipal and public-commitment material, the sizing argument and the recovery material each became a standalone article, leaving this one on tenant credit and the present value case. The evidence is unchanged and the findings that bear on the carry model, particularly the abatement interaction, remain here.

September 14, 2026, revision 5. Corrected the recovery section. An earlier version of this revision stated that no transfer mechanism exists for a large load interconnection request. That was wrong. ERCOT's Batch Zero readiness guidance expressly contemplates transfer of the request to a new entity without loss of Batch Zero classification, subject to the new entity meeting eligibility criteria. The correct distinction is that capacity reassignment is prohibited under the adopted rule while the request transfers with the project entity. Added the Graham transaction benchmark, market re-leasing conditions, and the alternative-use ranking. Corrected the financial security discussion to reflect the tiered retention in the adopted order, including the 50 percent tier for batch zero loads not meeting the Planning Guide maturity criteria, which the prior version misattributed to false information.

September 14, 2026, revision 4. Added a recovery section. Added the hedge mismatch exposure, framed two-sided after checking the rate environment. Clarified that the $50,000 per MW figure is refundable security rather than a fee, distinct from the statutory flat study fee. Condensed the municipal and incentive material.

September 14, 2026, revision 3. Replaced the carry-to-remedy ratio with a present value comparison net of residual. The prior version compared one year of carry to the entire damages claim, which mismatched a recurring cost against a one-time recovery and ignored deferred revenue. The corrected figure is materially larger, not smaller. Decomposed carry into cost of capital, property tax, opex and insurance, and flagged the abatement interaction on the tax line. Added retrofit premium, queue-slot loss and cost-of-preserved-optionality figures to the counterweight section. Condensed the counterparty diagnostic section to the counter-offer signal.

September 14, 2026, revision 2. Corrected the ERCOT phased-milestone trigger from six months to twenty-four, and corrected the security retention from 80 percent to 20 percent, both per Staff's Recommended Adoption Order in PUCT Project 58481. Added executed Chapter 380 and 381 agreement evidence on recapture range and phase scoping. Corrected an earlier claim that reserved water capacity generally carries a standing charge: the executed Texas data center agreements reviewed contain no take-or-pay.

September 14, 2026, revision 1. Initial publication.

---