Observational AI · Field Notes
Notes on performance, ROI, risk, and responsibility in enterprise AI -- written for the people who sign the purchase orders.
Field Notes by email
One email per article, a few a month at most. Unsubscribe in one click.
AWS drew four operating models for inference. Google Cloud and Azure have the same four; only the product names, the silicon, the billing details, and the depth of the Kubernetes tier differ.
Token-priced APIs on the left, self-run GPU clusters on the right, two managed stops between. A cross-cloud crosswalk and two worked cost examples.