What an AI answer costs, and when a GPU of your own pays for itself
One help-centre answer costs $0.0005 to $0.05 depending on the model. A GPU of your own beats Claude Sonnet 5 at about 196 answers an hour, but GPT-6 Luna at 3,918.
7 min read
Topic
Cost & FinOps
Where the money actually goes in a data platform, and the levers that move it: storage layout, compaction, instance shape, and retention nobody wanted.
The interesting part of platform cost is not the invoice total; it is the workload behavior that produced it. File sizes, rewrite frequency, storage layout, idle capacity, and retention policy usually matter more than a discount negotiation. These notes show how I trace a bill back to engineering choices, change one lever at a time, and keep the performance and recovery consequences visible beside the savings.
2 posts
All writingPer scan or per hour: the number that decides an analytics bill
Per-scan billing beats per-hour warehouses until an hour of compute scans about 1.4 to 3.4 TB. The break-even from list prices, and how to find your own number.
7 min read