Persistent Disk pricing is provisioned GiB per month by disk type, billed by the second from creation to deletion whether an instance is attached or not; Hyperdisk comes with two further dimensions, the provisioned IOPS and the provisioned throughput per month above a baseline, regional disks cost double, Local SSD costs per GiB on the instance, and snapshots pay per GiB and month of their compressed, incremental size, with rates of their own for archive and for retrieval. On the bill, the disks nobody needs any more dominate, because they keep running after the instance is deleted, and the SSD types provisioned for workloads Balanced would have carried.

Pricing dimensions

Google charges block storage by provisioned size, not by used size, and on Hyperdisk additionally by provisioned performance; every combination of type and region has its own SKU, and the prices differ per region.

DimensionUnitWhat drives itOptional
Persistent Disk Standard, Balanced, SSDGiB per month provisioned, by the second; I/O included in the price; performance grows with sizeEvery disk, attached or notNo
Persistent Disk ExtremeGiB per month plus provisioned IOPS per monthDisks with their own IOPS settingYes
Regional Persistent DisksDouble the zonal rate per GiB, replicated synchronously in two zonesHigh availability at disk levelYes
Hyperdisk BalancedGiB per month plus IOPS and MBps per month above the baseline of 3,000 IOPS and 140 MBps; the High Availability variant at double the rateThe newer families N4, C4 and others that attach only HyperdiskNo on those families
Hyperdisk Throughput, Extreme, MLThroughput: GiB plus MBps per month; Extreme: GiB plus IOPS per month; ML: GiB plus MBps per monthDisks for throughput, for IOPS, for modelsYes
Hyperdisk Storage PoolsThe pool's provisioned capacity and performance per month, Standard or Advanced, instead of per diskMany disks with thin provisioningYes
Local SSDGiB per month on the instance, with columns of its own for Spot and commitments; lost on stopInstances with local scratch storageYes
Standard snapshotsGiB per month of the compressed, incremental size, regional or multi-regional, at least one hour; network charges on creation and restore across location boundariesEvery snapshot as long as it exists, after the disk's deletion tooYes
Archive snapshotsGiB per month at the lower rate, at least 90 days, plus retrieval per GiB on restoreRarely needed backupsYes
Instant snapshotsGiB per month at the disk type's rate, in the disk's zoneFast local restore pointsYes
Custom images and machine imagesGiB per month of the image size at the storage locationImported and created imagesYes
Purchase optionLocal SSD under compute commitments; Persistent Disk and Hyperdisk without a commitment of their ownYes (Google Cloud committed use discounts)

Size is the price: a Persistent Disk pays every provisioned GiB, written or empty, and on Standard, Balanced and SSD the performance is bound to the size, so a disk is often provisioned larger than the data needs to get IOPS. Hyperdisk separates the dimensions: capacity costs per GiB, performance per provisioned IOPS and MBps per month, and the baseline of 3,000 IOPS and 140 MBps is included on the Balanced type; whoever orders more pays it every hour whether the application draws on it or not. The newer machine families N4 and C4 attach only Hyperdisk, so a change of family changes the disk line too. Snapshots pay their compressed difference to the previous snapshot, not the disk size, and live on independently of the disk; an archive snapshot costs less per month but at least 90 days and a retrieval per GiB. The mechanics across all three providers are in block, object and file storage; what storage tiers and lifecycles do, in storage tiers and lifecycle.

How it appears in your billing data

In the billing export every disk row carries service.description Compute Engine and a sku.description per type and region: Storage PD Capacity in Frankfurt for Standard, Balanced PD Capacity in Frankfurt, SSD backed PD Capacity in Frankfurt, Regional Balanced PD Capacity in Frankfurt, Hyperdisk Balanced Capacity in Frankfurt with Hyperdisk Balanced IOPS in Frankfurt and Hyperdisk Balanced Throughput in Frankfurt as rows of their own, SSD backed Local SSD in Frankfurt for Local SSD, Storage PD Snapshot in Frankfurt for regional and Storage PD Snapshot for multi-regional snapshots, Archive Snapshot Storage and Archive Snapshot Retrieval, Storage Image for images; usage.unit is gibibyte-month for capacity and snapshots and IOPS- or MBps-month for performance, usage.amount the provisioned quantity over time. resource.name in the export with resource-level detail names the disk or snapshot, and location.zone the zone; without resource-level detail, disks and snapshots sit as a sum per SKU and project, with no link to the instance, because system_labels carry the instance's machine data, not the disk's. The disk's labels in labels are the attribution per team, and a disk without an instance often has none, because the label came with the instance.

Discounts sit in credits: Local SSD under a commitment carries credits.type COMMITTED_USAGE_DISCOUNT, and Persistent Disk and Hyperdisk have no sustained use discount and no commitment of their own, so their rows sit at list price. The export's structure is described in the Google Cloud billing export; in the FOCUS export, ServiceName carries Compute Engine, ServiceCategory Storage, SkuId the SKU and ChargeCategory Usage (FOCUS columns). Which disk hangs on no instance is not in the export but in the disk list with an empty users field; the search is described in idle and orphaned resources.

A diskA Compute Engine row: the SKUnames type and region, StoragePD, Balanced, SSD, Regionalor Hyperdisk CapacityHyperdisk: IOPS and throughputas rows of their ownAt list price: no sustaineduse discount and no commitmentfor Persistent Disk and HyperdiskLocal SSDUnder the compute commitment:a committed use discount credit
A disk is a Compute Engine row at list price with its type and region in the SKU, Hyperdisk performance in rows of its own, and only Local SSD under a commitment

Levers

Configuration: the first lever is the type, because SSD costs four times Standard and 1.7 times Balanced per GiB and Balanced suffices for most workloads; on N4 and C4, Hyperdisk Balanced is the default type and costs less per GiB than Balanced as long as the baseline of IOPS and throughput is enough. The second is the size, because every provisioned GiB pays and a disk does not shrink; whoever enlarged for performance moves to Hyperdisk, where performance is ordered separately. The third is the snapshots: a schedule with retention, archive for everything that stays longer than 90 days, and deleting the snapshots of deleted disks; the patterns are in storage optimisation.

most workloadsmeasured needfor moreyesWhich type?Balanced; on N4 and C4Hyperdisk Balanced, cheaperper GiB while its baselineof IOPS and throughput sufficesSSD, at four timesStandard per GiBThe size: every provisioned GiBpays and a disk never shrinksEnlarged onlyfor performance?Hyperdisk: performance orderedseparately, from measurement,because provisioned IOPSpay every hour
Balanced unless measured otherwise, the size as small as the data because a disk never shrinks, and performance ordered separately on Hyperdisk from measurement

Purchase option: none for Persistent Disk and Hyperdisk; Local SSD falls under the compute commitments and has Spot prices with a smaller discount than the machine.

Architecture: regional disks only where the application does not provide the availability itself, because they cost double; Local SSD for scratch, never for data that must survive a stop; and the performance on Hyperdisk from measurement, not from estimate, because provisioned IOPS pay every hour. What the instance costs on top is in Compute Engine pricing.

Where it goes wrong

  • The disk outlives the instance. A disk without an instance keeps paying every GiB. PD Capacity rows without matching core rows and the disk list with an empty users field show it; deletion on instance deletion is a setting.
  • SSD for a log. A pd-ssd for an application with few IOPS costs four times Standard. The disk's metrics show the IOPS used; Balanced or Hyperdisk Balanced suffice.
  • Provisioned large for IOPS. A 2 TiB Balanced disk for 100 GiB of data, because the IOPS come with the size. Hyperdisk separates capacity and performance; the bill falls to the data plus the ordered IOPS.
  • The ordered IOPS run empty. Hyperdisk Balanced with 20,000 IOPS pays 17,000 above the baseline every hour, at night too. The provisioning can be changed; the measurement shows the need.
  • Snapshots without end. A daily snapshot schedule without a retention rule grows with every change. A retention of days, archive for months, and deletion with the disk.
  • The archive snapshot was deleted after a week. Archive pays at least 90 days and per GiB on retrieval. Archive only for backups that stay.
  • The regional copy for development. A regional disk costs double, in environments without an availability requirement too. Zonal for everything but production.

Example

List prices in USD for Frankfurt (europe-west3), read on 2026-10-21 from the disk and image pricing page, per GiB-hour and converted to a month of 730 hours: Standard USD 0.000065753 per GiB-hour, USD 0.048 per GiB-month; Balanced USD 0.000164384, USD 0.12; SSD USD 0.000279452, USD 0.204; regional Balanced USD 0.000328767, USD 0.24; Hyperdisk Balanced USD 0.000128767 per GiB-hour, USD 0.094 per GiB-month, plus USD 0.000008219 per provisioned IOPS-hour, USD 0.006 per IOPS-month, and USD 0.000064384 per MBps-hour, USD 0.047 per MBps-month above the baseline; Hyperdisk Throughput USD 0.006 per GiB-month plus USD 0.295 per MBps-month; Local SSD USD 0.000131507 per GiB-hour, USD 0.096 per GiB-month; standard snapshots USD 0.000080822 per GiB-hour, USD 0.059 per GiB-month; archive snapshots USD 0.000030685, USD 0.0224 per GiB-month, retrieval USD 0.0224 per GiB; images USD 0.06 per GiB-month.

A Balanced disk of 500 GiB costs USD 60.00 a month, as SSD USD 102.00, as Standard USD 24.00, as a regional Balanced disk USD 120.00. The same need of 100 GiB of data with 6,000 IOPS costs USD 122.88 as a Balanced disk enlarged to 1 TiB for the IOPS, and as Hyperdisk Balanced with 100 GiB and 6,000 IOPS 9.40 plus 3,000 × 0.006 = USD 18.00, USD 27.40 together, because the first 3,000 IOPS are included. A daily snapshot schedule for the 500 GiB disk with 5 GiB of change a day and 30 days of retention holds about 650 GiB of compressed data and costs USD 38.35 a month; the same data in archive USD 14.56, plus USD 11.20 of retrieval on a restore of 500 GiB. Fifty orphaned disks of 100 GiB Balanced cost USD 600.00 a month. The figures show that Hyperdisk cuts the bill for IOPS-driven disks to a quarter, that the type multiplies the price fourfold, and that snapshots and orphaned disks are the lines that keep running without an instance.

Sources

Read on 2026-10-21. Prices: disk and image pricing. Service: storage options for Compute Engine, Hyperdisk. Billing data: structure of the standard Cloud Billing data export.