Debt-fueled expansion under pressure
The major AI labs have spent astronomical sums on GPUs and data centers, much of it financed with debt. Now lenders are starting to question whether those investments will pay off.
According to The AI trade now runs on borrowed money, the big AI providers are far from recouping their investments. Open-weight models are only months behind, and the giants have barely managed to build a moat. Lenders are beginning to demand higher returns to keep financing AI development.
Consequences for hosting economics
As lenders reprice the risk, it has direct consequences for companies dependent on API providers:
- Higher API prices: To cover the cost of pricier capital, API prices get pushed upward
- Less predictability: Debt-financed providers can change prices and terms quickly
- Open weights become relatively cheaper: Self-hosting open-weight models becomes more attractive as API providers have to pay more for capital
The portfolio approach wins
As Zentera points out, not every layer of a model portfolio needs to be rented. The familiar, high-volume workloads that make up most enterprise token volume — such as classification, extraction, code completion, and document processing — are strong candidates for self-hosting.
A file you host
As one recent analysis puts it: "An open-weight model is a file you host." That's a technical reality with major economic implications once debt-financed API providers come under pressure.
What does this mean for Danish companies?
For Danish companies, it means that dependency on debt-financed API providers is becoming risky. Self-hosting open-weight models offers:
- Predictable costs, independent of someone else's creditworthiness
- Control over data and infrastructure within the EU
- The ability to match models to tasks without API constraints
At Liviate, we help Danish companies host open-weight models on EU infrastructure.