Barclays says AI model companies pay roughly $35–40 of every $100 in revenue as
inference compute fees to the three major cloud providers — Amazon Web Services
(AWS), Microsoft Azure and Google Cloud Platform (GCP). Cloud vendors retain
about $10–20 in operating profit from that, implying roughly 35–45% operating
margins on compute fees. Barclays finds AI labs' paid inference margins jumped
from low-double-digit levels in 2025 to about 50–65%+ in 2026, with adjusted
gross margins up 30–50 percentage points YoY. Analysts say actual margins could
be higher but expect them to gradually retreat as frontier-model competition
intensifies and compute supply expands.
