Glossary
Cloud cost management
Cloud cost management is the practice of measuring, attributing, forecasting and reducing what an organization spends on cloud services. It links the provider's billing model to the teams and workloads that generate each charge.
Why cloud cost management matters at petabyte scale
Compute costs follow how long instances run, and they fall when instances are switched off. Storage costs behave differently. They follow how much data is held, how often it is read and where it moves, and data seldom shrinks: retention rules, backup copies, logs and AI training sets add capacity month after month. A per-gigabyte rate that is a rounding error at 100 TB is a major budget line at 10 PB. Read and transfer charges grow with use, so a storage bill can rise in a month when stored capacity stays flat.
How cloud storage is billed
Public cloud object storage is metered in several independent units, each with its own driver.
| Charge | Unit | What drives it |
|---|---|---|
| Capacity | GB-month, by storage class | Average volume held during the month |
| Requests | Per 1,000 operations | PUT, GET and LIST calls; material with many small objects |
| Retrieval | Per GB read | Reads from infrequent-access and archive classes |
| Data transfer | Per GB moved | Data leaving the provider or crossing regions |
| Minimum duration | Days per object | Objects deleted early from colder classes |
The arithmetic shows how quickly these add up. At a hypothetical $0.02 per GB-month, holding 1 PB (1,000,000 GB) costs $20,000 a month, or $240,000 a year, before a single read. Reading 10% of that petabyte out of the provider each month at $0.09 per GB adds 100,000 × 0.09 = $9,000 a month. Transfer out of a provider is covered under egress fees.
Allocation and the FinOps practice
Cost management starts with knowing who caused each charge. Providers support this through account hierarchies, resource tags and cost reports filtered by those tags. Untagged resources form a remainder that nobody owns. Two reporting models are common: showback reports consumption to each team without billing it, and chargeback moves the cost into the consuming team's budget. Either can be expressed as a unit cost, such as cost per terabyte stored or per customer served, which ties spending to business activity.
FinOps is the name the FinOps Foundation gives to this operating discipline. Its framework describes a repeating cycle of three phases: inform (visibility and allocation), optimize (reducing waste) and operate (tracking targets and policy). Spending limits and approval rules overlap with cloud governance.
Optimization levers and their trade-offs
- Commitments exchange a discount for an agreement to pay for a set level of use over one or three years.
- Class transitions move ageing data to cheaper classes, described under cloud storage tiering.
- Clean-up removes orphaned snapshots, unattached volumes, stale replicas and incomplete multipart uploads.
- Placement keeps compute near its data to avoid transfer charges, a consequence of data gravity.
Each lever carries its own cost. Moving data to an archive class lowers the capacity charge and raises retrieval charges, retrieval time and early-deletion exposure, so the saving depends entirely on how often the data is read.
What cloud cost management means for large-scale storage
Access patterns decide the bill more than capacity does. Two datasets of the same size can differ several times over in monthly cost if one is read constantly by analytics or AI pipelines and the other sits untouched. Storage teams forecasting cloud cost at scale therefore need read and transfer volumes per workload, which most capacity plans do not track.
AI workloads change the read profile. Training and retrieval pipelines read the same corpus repeatedly, sometimes from GPU capacity in another region or another cloud. A dataset priced as cold storage on the assumption of rare access can become one of the largest lines on the bill once a training cluster starts reading it every epoch.
Exit has a price that grows every month. Moving 5 PB out of a provider at a hypothetical $0.05 per GB costs 5,000,000 × 0.05 = $250,000 in transfer alone, and the figure rises with every petabyte added. That number sets a practical ceiling on how freely data can move between clouds or back on premises, and it belongs in any decision to place a large dataset.
Predictability matters as much as the total. Variable request and transfer charges make storage spend hard to budget a year ahead, which is one reason organizations with large, steadily growing datasets compare metered cloud pricing with owned capacity, where cost is set mainly by hardware, power, space and staff.
Storage cost with Scality RING
Scality RING is software-defined object and file storage that runs on standard x86 servers. On infrastructure an organization owns or leases, storage cost takes the form of hardware, software licences, power, space and operations, with no provider-metered request or transfer charges for reads inside that environment. Cost management in that model becomes a question of capacity planning and media choice. Scality ADI places hot, warm and cold data on different media under one policy-driven lifecycle, which Scality calls "power-aware media tiering".