London – On Tuesday, Nscale announced it has closed a $900m revolving credit facility, a standing pool of borrowing that will be used to accelerate the construction of data centers across the United States, Europe, and Asia-Pacific. The move comes just weeks after a $2bn Series C round in March that valued the company at $14.6bn, and it signals a notable shift: the race for AI compute capacity is now entering the debt-fueled phase.

Until now, the AI infrastructure sector had been fueled almost exclusively by equity, venture capital, and patient backers willing to bet on the explosive demand for GPUs and data centers. The entry of bank debt, through a revolving credit facility of this size, sends a precise message: expected cash flows are now seen as predictable enough to secure a loan. It’s no longer just backing an idea, but an industry with measurable costs and revenues—a transition from a speculative bet to an asset class treated almost like a utility.

What does this mean for those building and running Large Language Models (LLMs)? First, the massive expansion of data center capacity—Nscale is targeting three continents—promises to ease, in the medium term, the compute supply bottleneck that has defined the last two years. While this could reduce cloud GPU rental costs, it also makes the choice between cloud and on-premise deployment more nuanced. For organizations considering bringing LLMs in-house, greater availability of rented infrastructure may make cloud computing more competitive on a Total Cost of Ownership (TCO) basis, but it doesn’t solve the sovereignty and operational control equation. In a landscape where a single player accumulates billions in debt to build facilities that will host third-party workloads, infrastructure concentration becomes a variable to watch—especially in Europe, where regulations like GDPR impose strict data residency requirements.

Nscale’s bet is large: transforming from a startup into a global AI utility, a trajectory reminiscent of telecoms in the early 2000s. Debt markets have responded positively, but the real test will be filling those facilities with customers willing to pay a premium for compute proximity. Either way, the operation sets a benchmark: from now on, the growth of AI infrastructure is no longer measured only in billion-dollar rounds, but also in credit lines.