Source: Applied Digital has asked tenants like Oracle and Meta for billions in guarantees since at least last year to secure power before finalizing leases
Tech giants chasing the infinite promise of AI are finding themselves increasingly on the hook for something far earthier — the electrical grid.
Context & Ripple Effects
AI capacity commitments have already outgrown straightforward equipment purchasing: OpenAI's more than $1T in computing-power deals raised funding questions, while Oracle used special-purpose vehicles to shift data-center construction debt off its balance sheet.
Power has become the binding input in that buildout. Big Tech's growing role in AI-driven electricity demand put grid costs and capacity under scrutiny, and banks later struggled to distribute loans tied to Oracle-leased data centers.
First-order effects
- Oracle and Meta are being asked to back Applied Digital's power procurement with multibillion-dollar guarantees before leases are finalized, putting their credit behind capacity they intend to use.
- Applied Digital gains a route to secure electricity earlier in the development cycle by tying its power commitments to the credit of prospective hyperscale tenants.
Second-order effects
- Tenant guarantees make Oracle's and Meta's balance sheets part of the financing case for AI data-center power, rather than leaving developers and their lenders to underwrite demand solely through future leases.
- For lenders facing the risk highlighted by difficulty distributing Oracle-linked data-center loans, explicit tenant support can become a more important condition for financing power-intensive projects.
Third-order effects
- If this structure spreads, AI capacity contracts will increasingly bundle compute, real estate, electricity and tenant credit into a single infrastructure-finance stack.
- The model also concentrates long-duration grid and project risk at the largest AI buyers, strengthening their advantage over customers unable to provide comparable guarantees.
The trend: AI infrastructure is being financed less as standalone data-center real estate and more as utility-scale capacity backed by hyperscaler credit.