Private Credit Just Wrote a $1.3 Billion Check to an Anthropic Data Center — and Nobody's Asking the Obvious Question

Private Credit Just Wrote a $1.3 Billion Check to an Anthropic Data Center — and Nobody's Asking the Obvious Question

Let me tell you something that's been sitting wrong with me all week. On Wednesday, Bloomberg reported that Eagle Point Credit Management — a private credit shop best known for CLOs — is writing a roughly $1.3 billion loan for a sprawling AI data center in Texas tied to Anthropic. That's not the headline. The headline is that this loan exists at all. Because this isn't one lender taking a flier on one building. This is the latest brick in a financing stack so deep that if you stand at the bottom and look up, you can't see daylight.

I've been running hosting infrastructure for over a decade, and I've watched this industry finance itself every which way. But I have never seen a financing structure like the one this project is sitting on. So let's pull it apart — because what's happening in Hubbard, Texas, tells you more about where this AI buildout is headed than any earnings call or capex number ever will.

The Loan Nobody's Talking About

Let me be precise about what happened. Eagle Point Credit Management LLC is providing roughly $1.3 billion in private credit for a data center campus in Texas that's tied to Anthropic. Bloomberg called it "one of the latest jumbo financings fueling the artificial intelligence boom." That's a polite way of saying the AI buildout just found another piggy bank.

Now, Eagle Point isn't a bank. It's a credit manager — the kind of firm that buys and structures debt, the kind of shop that a decade ago would have been buying collateralized loan obligations, not funding gigawatt-scale concrete. And that's exactly the point. The banks are already in this deal up to their eyeballs. The private credit market is now piling in behind them. And somewhere above both of them sits Google's balance sheet, holding the umbrella.

The Financing Stack Nobody Can See From the Road

Here's the full picture, and I want you to sit with it. The developer is Nexus Data Centers. The site is in Hubbard, Texas. Back on July 30, CNBC and the Wall Street Journal reported Nexus was in advanced talks for about $15 billion of financing, with Morgan Stanley leading the bank group. The package includes a roughly 1.6-gigawatt on-site natural-gas power plant — not power from the grid, power on the property, because the grid can't be trusted to show up.

And here's the part that should make every independent hosting provider reading this sit up straight. Google is guaranteeing Anthropic's lease and power-payment obligations if Anthropic defaults. In exchange, Google is expected to take roughly 20 percent equity in the data-center and power project. Anthropic plans to fill the site with TPUs co-designed by Google and Broadcom, with the chip costs running on a separate vendor-financing track.

Read that again slowly. The lender of last resort isn't a lender. It's a hyperscaler. Google is not writing a $15 billion check — it's putting its investment-grade balance sheet behind a tenant so the banks will fund the concrete and the gas turbines. That's not an investment. That's a credit enhancement wearing a trench coat.

Reading One — The Bull Case: The Money Found the Projects

Let me give the bulls their due, because they're not wrong about everything. The private credit market writing $1.3 billion checks means institutional money believes AI compute demand is real and cash-generative. That's the same thesis that produced CoreWeave's record quarter and the $60 billion order backlogs at hardware suppliers. When private credit — the most return-obsessed, yield-hungry capital on earth — is willing to fund a data center in rural Texas, that's not a sign the buildout is fake. That's a sign the buildout prints money, or at least that smart money thinks it will.

And look at the structure from Google's side. Google gets a captive customer buying TPUs on a vendor-financing track, a 20 percent equity stake in the power and data-center project, and a guaranteed offtake for its own chip ecosystem. That's not charity. That's vertical integration funded by other people's balance sheets. From Google's seat, this is a beautiful deal. The risk sits with the lenders and the ratepayers, not with the hyperscaler.

Reading Two — The Warning: Google's Balance Sheet Is Wearing a Trench Coat

But here's the thing about credit enhancement. It doesn't make risk disappear. It moves it around, and usually it moves it somewhere nobody's looking. This deal is a triple stack: private credit at the bottom, bank loans in the middle, and a hyperscaler's investment-grade rating on top. Every layer believes the layer above it will catch the fall. And the layer at the very top — Google — believes Anthropic will keep paying rent forever.

Now, I've covered this pattern before. Nvidia just signed up Apollo, BlackRock, Blackstone, Brookfield, Goldman Sachs and KKR to turn GPUs into loan collateral — a "market for credit backed by NVIDIA compute," as Goldman put it. The banks are at $334.5 billion in data-center debt issuance year to date, with an estimated $1.65 trillion in hidden off-balance-sheet obligations sitting underneath it. And here's the tell: in February, Blue Owl couldn't arrange financing for a $4 billion CoreWeave data center in Lancaster, Pennsylvania. Lenders weren't sold. Six months later, private credit is writing $1.3 billion for a project whose tenant's rent is guaranteed by Google. The difference between those two deals isn't the project quality. It's the backstop. The market didn't get more confident. It got more collateral.

The Repossession Wave — What Happens When These Loans Go Bad

Which brings me to the part of this story that nobody in the financing stack wants to talk about: what happens when the loan goes bad. There's a video making the rounds — The Real BPEarthWatch put it up on August 17 — called "The Repossession Wave: Why lenders are taking back physical control of data centers and Nvidia chips." And that title is doing a lot of heavy lifting, because it describes the endgame of everything I just walked you through.

Here's the ugly math. These loans are collateralized by GPUs and by the buildings themselves. A GPU has an effective depreciation curve of roughly 18 months before the next generation makes it obsolete. The loans secured against those GPUs run 36 months or longer. Do that math. By the time a lender actually repossesses a fleet of Nvidia chips, the collateral has lost half its value — and the next-generation chips that replaced it are already on someone else's balance sheet. That's not a repossession. That's a discount liquidation with extra steps.

And when the collateral is a data center building — a building full of power infrastructure, cooling, and fiber that was custom-built for one tenant — repossessing it doesn't give you a marketable asset. It gives you a very expensive monument to the fact that nobody else wants a gigawatt-scale facility in a specific county in Texas. Lenders are already starting to take physical control of facilities and hardware. Watch that trend, because it's the canary. When the people who lend the money start collecting the machines, the financing party is over.

What This Means for Independent Hosting Providers

So what does a $1.3 billion private credit loan in Texas mean for somebody running a colo or a hosting business with real customers and a real P&L? More than you'd think. Let me give it to you straight.

First, watch the credit signals, not the headlines. When a hyperscaler's backstop gets tested, or a private credit lender starts marketing repossessed GPU inventory, that's your early warning. It'll show up in CDS spreads and secondary GPU markets months before it shows up in the news cycle. Set up a watchlist. The repossession wave is your leading indicator.

Second, lock your equipment and power contracts now. Every layer of this financing stack is competing for the same transformers, the same turbines, the same grid capacity. When stressed borrowers start defaulting, supply chains don't get looser — they get tighter, because the collateral liquidations and the new builds are pulling from the same pool. The window to lock long-term pricing is open right now. It will not stay open.

Third, position as the capital-light alternative. This is the whole ballgame. The giant borrowers are financing gigawatt campuses with 36-month loans against 18-month hardware. You don't have to do that. Your advantage isn't scale — it's that you don't need Google's balance sheet to guarantee your rent. Independent hosting is the only part of this industry that doesn't require a credit stack to exist. Sell that. It's real, and it's becoming rarer by the week.

Fourth, read your cyber insurance and financing fine print. If you carry debt against your own hardware, check what happens on repossession and power-quality events. The market is tightening the exclusions even as it loosens the terms. The fine print is where the next generation of horror stories lives.

The Bottom Line

Here's what I actually believe. The AI buildout is not a fraud and it's not a bubble in the 2008 sense. The demand is real. But the financing has left the rails of anything resembling traditional credit discipline. When a private credit firm funds a data center whose rent is guaranteed by a hyperscaler's balance sheet, what you're really looking at is the hyperscaler borrowing against its own rating through a series of increasingly creative middlemen. The concrete gets poured, the turbines get ordered, and the risk gets stacked so high that nobody at the bottom can see the top.

I've said it before and I'll say it again: this isn't a shortage, it's a famine — and now the famine has a repossession clause. The lenders are starting to collect the machines. When that wave breaks, the independent operators who stayed lean, stayed capital-light, and stayed real are going to be the only ones still standing with their own equipment and their own customers. Keep your balance sheet boring. It's about to be your best competitive advantage.

— Allan Ali, Founder

This article was produced with AI-assisted research and editorial support. Sources: Bloomberg (Aug 19, 2026), CNBC/WSJ (Jul 30, 2026), explainx.ai (Aug 1, 2026), Business Insider (Feb 2026), The Real BPEarthWatch (Aug 17, 2026).