Have Your Say

About data centers – long, but explains a lot:

by Doug Casey

The largest infrastructure buildout in human history is currently underway, financed almost entirely by private capital, justified almost entirely by a commercial promise the underlying numbers clearly do not support.
What I do want to point out is that the product being built is exactly the kind of centralised data infrastructure a surveillance apparatus would need, if one ever wanted to be built on top of it. That alone should concern any thinking man (not that there are many of those left, going by the general public discourse).
Nobody has to plan this outcome for it to happen. The state doesn’t need to build the panopticon. It just needs to wait for the market to build one, misprice it, and be sitting there, insolvent, when someone with the authority to nationalise, regulate, or “partner” shows up.
Capitalism is extremely good at mobilising enormous pools of capital toward a physical build. It has never once been good at asking who inherits the asset once the company that built it goes bust (which is exactly what happens once the economics catches up with the narrative).

How You Get a Business Community to Build What the State WantsYou don’t need a five-year plan or a ministry of construction for this. You need three things: cheap capital, a rule change that makes the capital cheaper still, and a story compelling enough that opting out feels like the crazy move. The media pushing the narrative helps too, obviously.
Take the rule changes. In June, SpaceX went public on the Nasdaq in the largest IPO in history, debuting north of $1.7 trillion and blowing past $2 trillion within weeks — the sixth-largest publicly traded company in the United States. Big hyped listing… fine, we’ve seen those before. What was unusual was what happened to the rulebook to make room for it.

The original financing plan for the AI data-centre buildout ran through private markets — PE funds, private credit, sovereign co-investment, asset-backed paper sold to insurers hunting for yield. Patient money, hard assets, contracted revenue on top. At least, that was the pitch two years ago.
Except, as you’ll know if you’ve been reading this fine publication for more than five minutes, private equity and credit is quietly blowing itself up, so that particular dog no longer hunts.
Funds can’t return capital to their LPs because exits have dried up. Dry powder sits uninvested because valuations won’t clear. Fundraising cycles are stretching on for years longer than sponsors modelled. The asset class simply doesn’t have the horsepower left to keep funding a buildout of this size on its own.
So everything moved onto public markets and corporate balance sheets, because that’s the only pool of capital left deep enough to keep the spend going.
Hyperscaler capex went from roughly $150 billion in 2023 to an estimated $725 billion this year, heading toward $1.1 trillion in 2027. At that run rate, aggregate capex crosses 100% of operating cash flow — meaning every marginal dollar of new data-centre capacity is now funded by debt or equity, not cash the businesses actually made. That, my friends, is nuckin’ futs!

Behind all of it sits the actual borrower whose habits are driving the whole show: OpenAI, whose take-or-pay compute commitments run into the hundreds of billions of dollars against an operating business that haemorrhages cash. Those commitments get serviced the way a subprime mortgage got serviced in 2005 — not out of income, but by finding the next sucker willing to fund the following round at a markup over the last one.
When OpenAI’s valuation jumped from roughly $86 billion in early 2024 to a reported $852 billion by this spring, it was the step-up between rounds keeping the whole machine solvent, not the cash flow.
That step-up, by the way, has been shrinking with every round — from north of 1.8x down to about 1.23x implied by the company’s own reported IPO target. “Implied,” not real, which is not that much different from “guessed.”
The IPO, whenever it actually happens, will be the refinancing of last resort… and the fact that it’s already slipped from 2026 into 2027 tells you the mark can’t clear at the level the structure needs.
Private equity got out… or, more accurately, ran out of rope doing a bunch of other stupid shit that’s now also blowing up.
So who’s left holding the bag? Retail investors, pension funds, index funds forced by their own rulebooks to buy whatever they’re told to buy.
The marginal financiers of a buildout whose central borrower can’t cover its own commitments out of income. Much like how 2008 was, at its core, a credit story masquerading itself as a real-estate story, this is a credit story in a tech wrapper.
What gets lost in both the hype and the doom is that the physical stuff doesn’t vanish when the financing snaps. The data centres stay standing, same as the fibre stayed in the ground after the dot-com bust. The chips stay racked.
What changes is the owner of the paper on top… and its price.
A credit event that wipes out equity holders doesn’t demolish a data centre. It merely hands the ownership to whoever’s got the balance sheet to buy distressed hard assets while everyone else is forced to sell.

A credit event that wipes out equity holders doesn’t demolish a data centre. It merely hands the ownership to whoever’s got the balance sheet to buy distressed hard assets while everyone else is forced to sell.
Who’s Left Holding the Bag?Now, ask yourself why this is being built in the first place… and who’s going to end up owning it?
The way I see it, this is exactly the infrastructure you’d want if you were quietly assembling a techno-feudal surveillance state… and the mechanics of a credit cycle breaking are precisely what let that transfer happen in the first place.
Private capital took the early risk, is currently eating the mark-to-market losses, and will most likely eat the realised losses too when the refinancing chain finally snaps. Retail and index money — forced into the trade by rule changes, remember — are the terminal bagholders in the “refinancing of last resort” story, because there’s no deeper pool of capital left behind them.
But the actual data centres — the compute, the storage, the fibre, the physical capacity to hoover up and retain enormous quantities of personal and behavioural data — don’t disappear when the equity does.
They get bought at a discount by whoever has the balance sheet and the patience to wait out the distress.
In the past, that “whoever” was the best-capitalised private players. At moments of real systemic stress, it was also government — equity stakes, bailouts, “strategic asset” designations, the whole emergency-intervention playbook that turns a private failure into a public asset overnight.
None of this requires conspiracy. The state just needs to wait for the leverage to catch up with everyone. That’s the plan, as far as I can tell — and it’s an option the state doesn’t even have to pay for up front. It just waits, the way it always has, for someone else’s bad debt to become someone else’s national security asset.