AI
Infrastructure Contract Economics: Prepayment + Debt Model
AI infrastructure companies (GPU clouds, data centers) are securing
hyperscaler contracts with a distinctive financing model: large
prepayments from customers (20-30% of TCV) plus debt financing for
GPU/infrastructure CapEx. This creates a unit economics pattern distinct
from traditional SaaS.
Evidence
- IREN Microsoft contract: $9.7B TCV / 5 years, $1.9B prepayment +
$3.6B debt at <6%, expected 85% EBITDA margin at scale
- Pattern emerging across AI infrastructure: customer prepayments
de-risk the debt, creating leveraged returns on infrastructure
deployment
- Bert (Q2 FY26 earnings-review): Goldman + JPMorgan willingness to
lend $3.6B at <6% secured by GPU assets + Microsoft cash flows is
itself third-party validation of contract quality; 95% of $5.8B GPU
CapEx covered by prepayment + facility combined
- NBIS Q4 FY25: Deferred revenue of $1.577B = 6.9× Q4 quarterly
revenue. ARR of $1.25B = 5.5× Q4 annualised revenue. With $1.9B of
pre-contracted but unrecognised revenue (ARR-revenue gap + deferred
balance), FY26 revenue pipeline was essentially fully visible at
year-end FY25. Quarterly revenue beat/miss vs consensus (-6.2%) was
immaterial noise; ARR was the only metric that mattered.
Implication
For AI infrastructure companies, evaluate: (1) prepayment coverage of
CapEx, (2) debt service vs contracted revenue, (3) ramp schedule to
positive FCF, (4) counterparty credit quality. Traditional SaaS metrics
(ARR, NDR) are less relevant than contract economics and utilization
rates. Additional signal (NBIS): When deferred revenue
exceeds ~3× quarterly revenue, treat reported revenue beats/misses as
noise — the economically meaningful metric is ARR and the deferred
revenue balance. Track the ratio of deferred revenue to trailing
quarterly revenue as the primary demand visibility indicator for
pre-scale infrastructure buildout companies.