Have you ever watched a market treat a “can’t miss” build-out like it suddenly has a limp? That is the feeling around a huge New Mexico campus meant to feed the next wave of artificial intelligence. Loans tied to that site, once treated as a straightforward way to get steel in the ground, are now changing hands in a range that credit desks call stressed. Not default. Not a fire sale. Just a price that says lenders want more yield for the same story they bought last year.
Why These Project Loans Suddenly Matter
The package is about $18 billion. It was raised late last year to start construction on a 1,400-acre campus in Doña Ana County, often discussed as Project Jupiter. The tenant story is familiar: cloud capacity leased to a major software firm that has a large compute agreement with a leading AI lab. Banks underwrote the debt, expected to sell most of it, and keep a modest hold. That last part is the part that is not going to plan.
Syndicate desks have been quoting the paper around 89 to 91 cents on the dollar. In loan-land, par is 100. A few points off can be noise. Nine to eleven points off, with a stalled distribution, is a message. I’ve found that markets rarely shout on day one. They whisper through marks, then they raise their voice through inventory that will not leave the book.
A price below par is not a verdict of insolvency. It is a tax on uncertainty.
Perhaps the most interesting aspect is how quickly the conversation shifted from “how big can this get” to “who is actually holding the risk.” Construction still happens. Power still has to show up. Water still has to be explained to neighbors. And the tenant’s own bonds are not trading like fortress paper either. In my experience, that mix is how a single project becomes a proxy for an entire cycle.
What Stressed Pricing Actually Signals
People hear “stressed” and picture a funeral. That is sloppy. In leveraged loan and project finance slang, stressed usually means the market is asking for a fatter spread because something in the original underwriting looks less clean. Delays. Permits. Concentration. A weaker parent credit. Take your pick. All four are in the air here.
Think of it like a house that still has good bones but a surveyor just found a crack in the foundation and a neighbor who might sue over the driveway. You do not walk away. You renegotiate the price. That is what 90 cents is doing. It is a bid with conditions attached, even if nobody wrote those conditions on a term sheet this week.
- Secondary marks in the high 80s to low 90s imply a higher required return than launch.
- Stalled syndication means banks keep more exposure than risk committees like.
- Parent credit trading soft makes project paper harder to place, even when the lease looks solid on paper.
- Local opposition adds a timeline risk that models hate more than they hate a slightly higher coupon.
None of that proves cash will miss. It does prove that the buyer base got picky. Institutional loan funds have concentration limits. Insurance money wants predictability. Private credit can take complexity, but not always at last year’s spread. When all three groups pause at once, inventory sits. Sitting inventory is how a rumor becomes a mark, and a mark becomes a headline.
The Campus, The Desert, And The Power Problem
Doña Ana County is not a cartoon of empty land. It is a real place with water politics, air-quality rules, and a public that has watched big projects promise jobs and then argue about dust and pipelines. The campus is designed as a massive compute site. Early plans leaned on a large gas-turbine microgrid, on the order of a couple of gigawatts, so the load would not lean on household ratepayers the same way a grid-tied plant might.
Then the energy path got messy. A request to run a natural gas connection across state land ran into a block. Lawsuits and permitting fights over air quality added months. You can debate whether the objections are fair. Markets do not referee fairness. They price delay. A data hall without firm fuel is a warehouse with a very expensive roof.
Company materials have stressed closed-loop cooling and limited draw on local water. That argument matters in a dry state. It has not ended the argument. I’ve sat through enough community meetings in other counties to know the pattern: engineers talk closed loops, residents talk trucks, haze, and whether the next drought will make yesterday’s study look cute. The gap between those two rooms is where political risk lives.
How The Financing Was Built
This is not a simple corporate revolver. The structure is closer to project finance with a famous tenant. Special purpose vehicles own or develop the boxes. Equity partners put in a slice. Banks write a large construction or mini-perm loan. The software firm leases the capacity. Cash from that lease is supposed to service the debt. On a whiteboard, it is elegant. In a credit committee, it is a stack of assumptions.
Those assumptions include on-time delivery, available power, a tenant that keeps paying, and a secondary market that will take the paper after the banks do the heavy lift. Break one assumption and the model still works, just with a wider spread. Break two and the syndicate starts calling accounts that said “maybe later.” Break three and you get quotes in the 90 handle and a lot of polite emails that never turn into tickets.
| Piece of the deal | What it was supposed to do | Where friction showed up |
| Bank consortium loan | Fund early construction at scale | Harder to distribute after launch |
| Equity partner capital | Absorb first-loss and sponsor the SPV | Market more cautious on next sites |
| Long tenant lease | Anchor cash flow for lenders | Tenant leverage and capex draw scrutiny |
| On-site power plan | Keep the campus independent of local rates | Pipeline and permit delays |
Similar packages have been used for other campuses in Texas and Wisconsin, and for other partner sites. The New Mexico slice is not the only large ticket. It is the one now wearing a public mark that looks tired. That visibility matters because credit is social. Once one deal prints soft, the next one has to answer for it in the first five minutes of a call.
The Tenant Balance Sheet In The Background
Here is the part some project-finance folks would rather skip. Even if the loan is ring-fenced, buyers still look through the fence. The tenant has been borrowing heavily to fund an AI infrastructure sprint. Rating agencies have already tightened the tone. One major shop cut the corporate rating in mid-year, leaving it a single step above speculative grade. That is still investment grade. It is also a small cushion if anything else goes wrong.
Fair-value marks on the company’s own long-term notes have, at points, implied a deeper discount than the project loans. That comparison is easy to misuse. Corporate bonds and construction loans are not twins. Still, the direction is the same: the market wants more compensation for Oracle-linked duration. When the parent trades heavy, the lease that was supposed to be the comfort blanket starts to look like a concentration problem.
Capex guidance in the tens of billions for the coming fiscal year does not help the vibe. Customer prepayments and remaining performance obligations can offset some of that, and bulls will recite those numbers until the room goes quiet. Bears will answer with a simple question. What if the spend stays front-loaded and the cash from customers arrives later than the model says? That question is doing a lot of work in loan pricing right now.
The AI cycle can still be real and the financing can still be tight. Those two facts are not enemies.
Banks Stuck With More Paper Than Planned
Syndication is a relay race. The arranging banks take the baton, run the first lap, and hand it to funds, insurers, and other lenders. If the handoff fails, the first runners keep jogging with extra weight. That is what people close to the process have described. Names in the group have included large European and U.S. firms that expected a cleaner sell-down.
Why would a sophisticated desk misjudge demand? Sometimes they do not. Sometimes demand exists at a price the sponsor will not accept, so the loan sits in a gray zone: not rewritten, not freely trading, just marked. Sometimes the buyer list is full of Oracle exposure already from other campuses. Concentration limits are boring until they stop a $500 million clip from clearing. Then they are the whole story.
There is also a human factor. Risk officers who blessed AI infrastructure last autumn are now being asked whether they want another helping. After a year of headlines about power shortages, chip supply, and circular vendor financing, “another helping” is a tougher sell. I do not blame them. I also do not think this automatically means the campus dies. It means the next dollar is more expensive, and expensive dollars change behavior.
Local Pushback Is Not A Footnote
Investors far from the Rio Grande can treat community opposition as color. That is a mistake. Permits are a critical path item. Air quality litigation can freeze equipment orders. A blocked pipeline can force a redesign toward other fuels or a slower phased build. Politics can add a moratorium threat, even if a moratorium never arrives. Markets hate optionality that only runs one way.
Polls and campaign talk in the state have included sharp skepticism toward large compute sites. Whether that hardens into policy is a separate question. The mere chance of a pause is enough to widen a spread. Lenders underwrite law, not vibes, but they also underwrite the time it takes for law to settle. Time is interest. Time is also a window in which another region can steal the next cluster of GPUs.
- Map every permit that still sits with a state or county desk.
- Separate water claims from power claims; they scare different voters.
- Ask whether the microgrid can be staged if the full gas plan slips.
- Price a six-month delay as a base case, not a shock case.
- Watch local elections the way you watch a rate decision.
That list is not activism. It is credit work. If you skip it, you are not underwriting a data center. You are underwriting a press release.
Is This An AI Bubble Signal Or Just One Tough Ticket?
Every cycle produces a poster child. This week, New Mexico loans are auditioning for the role. I would not give them the part yet. One soft mark does not prove that demand for compute is fake. It does prove that the financing stack grew faster than the buyer universe that is allowed to own it.
Look at the pattern across the sector. Hyperscalers and software firms are racing to lock power, land, and chips. They use leases, joint ventures, and off-balance vehicles so the headline leverage looks neater than the economic leverage. Rating agencies notice. Banks notice. Eventually the public loan market notices. When it does, the first projects that combine size, delay, and a stretched tenant get marked first. That is textbook, not prophecy.
Could this still snowball? Sure. If the next Texas or Midwest package has to print wider, equity partners will demand thicker protections. If equity partners hesitate, sponsors slow sites. If sites slow, the AI lab that needed the capacity looks for another landlord. That chain is possible. It is not guaranteed. Cycles usually break at the weakest joint, and right now the weakest joint is not demand for tokens. It is the patience of people who have to fund the buildings.
Rough mental model: Demand for compute = still loud Supply of power = tight and local Supply of credit = narrower than 2025 Political consent = no longer automatic
What Equity Investors Should Actually Watch
Stock traders love a clean narrative. Credit rarely offers one. If you own the software name, the loan mark is a mood indicator more than a cash-flow print. Watch three things that do show up in reported numbers over time: interest expense, lease commitments, and the gap between capex and customer cash. If those three move the wrong way together, the multiple compresses even if revenue keeps climbing.
If you own the banks in the syndicate, the question is simpler. How much of this paper is still on the warehouse line, and at what mark? A few billion of retained loans will not sink a global balance sheet. A pattern of unsold AI tickets across several sponsors might change how those banks bid the next deal. That is how a project story becomes a sector funding tax.
If you own private credit funds that might buy the loans at a discount, congratulations. You live for this. Distressed-adjacent project paper with a still-paying tenant is the kind of thing that can work if you diligence the power plan like a forensic engineer. It can also trap you in a long workout if the campus becomes a political football. Pick your team carefully.
A Practical Framework For The Next Six Months
I like checklists more than slogans. Here is the one I would keep on a single page.
- Marks: Do quotes stabilize above 90, or do they drift into the mid-80s?
- Syndication: Does a real third-party bid appear, or do banks keep rolling holds?
- Permits: Is there a dated path for fuel and air approvals, or only statements of intent?
- Parent credit: Do spreads on the software firm tighten after earnings, or keep leaking?
- Capex versus cash: Do customer payments cover a rising share of build cost?
- Politics: Does state-level talk harden into a pause, or fade after the news cycle?
Two or three of those flashing yellow is manageable. Four flashing at once is how a “unique project issue” becomes a funding freeze for look-alike deals. You do not need a crystal ball. You need to count the yellow lights without lying to yourself about which ones you wished were green.
The Human Texture Behind A Giant Number
Eighteen billion dollars is an abstraction until you walk a desert site at dusk and see the graded pads. Workers want the jobs. Counties want the tax base. Neighbors want quiet nights and a water table that does not become a punchline. Lenders want a schedule that matches the model they sold to their own investment committee. Those wishes can live in the same county. They do not always live on the same calendar.
That is why I keep coming back to the price. Ninety cents is a crowd of people admitting, quietly, that the calendars do not match. Maybe they will match again after a redesign, a new fuel plan, or a round of community concessions. Maybe they will not, and the campus opens later, smaller, or under a rewritten lease. Either way, the loan market has already voted that last year’s certainty was a little too neat.
We have seen versions of this in energy, in telecom, in housing. Big visions meet local physics. Capital arrives early because missing the boom feels worse than overpaying. Then somebody has to live with the asset for ten years. The people who live with assets tend to be less romantic than the people who announce them.
What Would Change My Mind
Bull case first, because doom is cheap. A firm power solution that survives the land office. A visible sale of a large loan slice to long-only accounts at a tighter spread. A tenant print that shows customer cash arriving on schedule. A local compact that turns opposition into enforceable monitoring instead of a blockade. Any two of those would make 90 cents look like a buying opportunity rather than a warning flare.
Bear case, without theatrics. Another rating nudge on the tenant. A second campus package that cannot clear. A court date that freezes turbines. A political pledge that freezes new sites statewide. That mix would push this from “stressed loan” to “template for wider AI project spreads.” I hope that does not happen. Hope is not a hedge.
Credit markets are often early, rarely polite, and almost never impressed by a keynote.
The Quiet Lesson For The Whole Build-Out
The lesson is not “never build in the desert.” The lesson is that the AI supercycle, if that is what this is, now runs through loan desks, county hearings, and gas regulators as much as it runs through model benchmarks. Chips get the poetry. Transformers get the keynote. Debt gets the truth, usually a few months late and a few points cheap.
If you only remember one thing, remember this. A stressed mark on a flagship campus does not end the race for compute. It changes the cost of staying in the race. Higher cost means fewer copycat sites, tougher lease terms, and more questions for any company whose growth story assumes infinite willing lenders. That is not a eulogy. It is a price. And prices, unlike slogans, have to clear.
So keep an eye on the desert. Keep a sharper eye on the bids. The buildings may still rise. The interesting question, the one the 90-cent quote keeps asking, is who will own the risk when they do — and at what number they will finally admit they were paid to wait.
