The Sovereign AI Factory Doesn't Escape the Debt. It Concentrates It.

The pitch is seductive. Palantir and Nebius launched a "sovereign AI stack" this week: modular data centers on sites that already have power, so an enterprise can own its compute and its data instead of renting both from a hyperscaler.

Organizations need both the performance of large-scale AI infrastructure and control over their data and models. — Arkady Volozh

DigitalBridge's chief executive describes the same turn in different words, customers drifting away from the big clouds toward "integrated AI factories" they control themselves. ^[1] The promise is an escape from two things at once: dependence on the hyperscalers, and the grid that can't power them. Own your own power, own your own compute, and the bottleneck disappears. It doesn't. The escape route costs more than the thing it replaces. Renting compute from a hyperscaler means someone else carries the power problem. Building a sovereign factory means carrying it yourself. The administration now requires AI firms to secure their own energy, and the result is at least 82 proposed gas plants, plus a 20-year Chevron deal to build a 2.67-gigawatt private power plant for Microsoft in Texas. ^[2] That is not a cheaper way to buy compute. It is the most capital-intensive version of the buildout yet, concentrated in fewer and larger bets, each one a private power plant plus a data center plus the chips to fill it. And the bets are being placed at the worst possible moment on the revenue side. Token prices have fallen more than 50% from their summer peak to a record low of 97 cents. ^[3] MIT's Project NANDA finds 95% of enterprise generative AI pilots show no measurable P&L impact. ^[4] Morgan Stanley's arithmetic is the sharpest version of the problem: a fully optimized Nvidia data center costs $25 billion a year to rent and generates $23 billion in output. ^[5] The factory is being sold as the answer to a debt loop, and it is being financed with more debt than the loop ever required. The crowding out is already visible, and it has left the AI sector. AI-related borrowing now exceeds 40% of new long-term investment-grade corporate debt, according to Apollo. ^[6] The European Central Bank counts US tech giants at roughly 10% of gross euro zone bond issuance. ^[7] BNP Paribas found hyperscaler debt sales pushing up the cost of insuring unrelated high-quality borrowers, LVMH and BAE Systems among them. ^[8]

All high-quality credit competes with hyperscalers for capital. — Josh Farberow

AI capital spending has helped drive 30-year US Treasury yields to 5.28%, the highest since 2007. ^[9] A French luxury house and a British defense contractor did not build an AI factory. They are paying for one anyway. The contradiction is now visible inside the market itself. Credit default swap spreads for Broadcom, Nvidia, and Oracle widened sharply in August, while equity markets stayed elevated. ^[10] Two markets are looking at the same sovereign AI factory and rendering opposite verdicts. The equity market sees growth. The credit market sees a borrower. The escape route was supposed to reduce exposure to the debt loop. It concentrated it, and the concentration is now repricing the cost of capital for everyone else.