Privacy and preparation
Dataset secure delivery cost comparison: cloud, staging or drives
By SourceX Editorial · Reviewed by Noah Loul ·
10 sections
Short answer
Compare delivery methods by calendar first: network time equals size divided by sustained bandwidth times utilization. If that fits the window, direct cloud transfer or time-limited managed staging usually wins, with automatic integrity and access records. If not, encrypted drives win on time but add custody work. Then total the cost lines at your own rates and check which method produces the evidence you need.
Value drivers
Dataset secure delivery cost
Key takeaways
- Network time in hours is terabytes x 8,000,000 divided by usable megabits per second, divided by 3,600.
- Measured utilization and repeat frequency flip the decision more often than unit prices.
- Cloud transfer and managed staging produce machine-generated integrity records; drives need a custody log.
- Managed staging adds a copy and an expiring link that must both be closed after receipt.
- Leave prices as inputs from your provider's current price list and contracts, and check device availability before planning.
How do cloud transfer, managed staging and encrypted drives compare for dataset delivery?
Start from the calendar, then the evidence, then the cost. Compute how long the network transfer would take at the bandwidth you can actually sustain; if that fits the delivery window, cloud transfer or managed staging usually wins, because both leave machine-generated integrity and access records. If it does not fit, or the data sits on premises behind a slow link, encrypted drives win on time and you pay in custody work instead.
Between the two network options, the deciding question is who should control access during the handoff. Direct cloud transfer gives the recipient a scoped grant on your storage or theirs, with no intermediate copy. Managed staging puts a time-limited copy in a landing area, such as an SFTP endpoint or a staging bucket with expiring links, which suits recipients who cannot accept a cross-account grant but adds one more copy to track and delete.
The worksheet on this page turns that rule into inputs, formulas and three worked scenarios. It deliberately leaves unit prices as inputs: cloud and courier rates change, differ by region and contract, and the right numbers are the ones on your own provider's current price list.
What exactly counts as each delivery method?
Name the method precisely before costing it, because the security properties follow from where copies exist and who holds the keys, not from the label. The table defines the three methods as this worksheet uses them.
| Method | What moves | Copies created | Access control | Typical tools |
|---|---|---|---|---|
| Cloud transfer | Objects copied storage to storage over the network | Source plus recipient copy | Scoped cross-account grant or service role | AWS DataSync, Google Storage Transfer Service, AzCopy |
| Managed staging | Files placed in a landing area the recipient pulls from | Source, staging copy, recipient copy | Per-recipient account or expiring signed link | Managed SFTP such as AWS Transfer Family, presigned or SAS links |
| Encrypted drives, seller-owned | Hardware-encrypted drives shipped by courier | Source, drive copy, recipient copy | Drive PIN or key released separately | Self-encrypting drives with validated modules |
| Encrypted drives, provider device | Provider appliance filled on site and returned to the provider | Source, device copy, cloud copy | Unlock password or PIN held in the provider portal | Azure Data Box, Google Transfer Appliance |
Which inputs does the worksheet need?
Collect these inputs before comparing anything. Most mistakes in delivery planning come from a guessed bandwidth or a forgotten second copy, not from the arithmetic.
Measure sustained throughput with a real test copy of a few hundred gigabytes over the actual path, at the time of day the transfer would run. Line speed on a contract is a ceiling, not a forecast, which is why the worksheet carries a separate utilization factor.
Take V from the final manifest, not from a storage console estimate. Compression and file packaging can shrink or grow the package, and a corpus of millions of small files can move far slower than its size suggests, because per-file overhead dominates; if that applies, package small files into larger archives before costing the network options.
| Input | Symbol | How to get it | Note |
|---|---|---|---|
| Package size | V (TB, decimal) | Sum of manifest file sizes | Use decimal TB to match the formula |
| Sustained bandwidth | B (Mbps) | Measured test copy on the real path | Lowest link between source and destination |
| Utilization | u (0 to 1) | Share of B available without hurting operations | Shared office links rarely allow full use |
| Delivery window | W (days) | Agreement or project plan | Include verification time, not just copy time |
| Repeat deliveries | n per year | Agreement schedule | Recurring feeds favor network methods |
| Unit rates | r values | Your provider's current price pages and contracts | Egress, staging storage, service, courier, drives |
| Staff time | h (hours) | Estimate per method from the procedure steps | Packing, verifying, custody logging, cleanup |
How do you calculate transfer time and total cost?
Transfer time over the network is size divided by usable bandwidth. With V in decimal terabytes and B in megabits per second, one terabyte is 8,000,000 megabits, so network time in hours is V times 8,000,000, divided by B times u, divided by 3,600. Add verification time, because a managed transfer tool that re-reads data to check it can add a meaningful pass on large jobs.
Drive time is copy-out time plus transit plus copy-in time plus verification, where each copy time is V divided by the combined write or read speed of the drives used in parallel. Provider device services add ordering and return logistics that the provider controls.
Cost is the sum of line items, each a quantity times your current unit rate. Write the quantities in the worksheet and multiply by rates from your own price list.
- Network time (hours) = V x 8,000,000 / (B x u) / 3,600.
- Drive time (hours) = copy-out + transit + copy-in + verification, with copy time = V x 1,000,000 / combined MB per second / 3,600.
- Egress = V in GB x egress rate, charged to whichever party the agreement names.
- Staging storage = V in GB x storage rate x months the staging copy exists, plus request charges.
- Managed service = service hours or GB moved x the service's rate.
- Drives and courier = drives bought or rented, plus shipping both ways and insurance.
- Staff = hours per method x loaded hourly rate, including cleanup and deletion proof.
- Annualized = one delivery x n, since a method that is cheap once can be expensive monthly.
How do the three methods differ on security evidence?
Each method can be made secure; they differ in which evidence is produced automatically and which a person must create. Machine-produced evidence is usually stronger in a dispute because nobody typed it.
Network tools verify integrity as part of the job. AWS DataSync checksums data during transfer and can verify either only the transferred data or the full source and destination at the end of a run. Google's Storage Transfer Service validates checksums end to end when the source provides checksum metadata, and records a failure when they do not match. Provider devices also leave records: Azure Data Box produces a manifest with file sizes and checksums, a copy log on upload and chain of custody logs after erasure.
Expiry is where staging needs care. Amazon S3 presigned URLs can be valid for up to 7 days when created with the CLI or SDKs and up to 12 hours from the console, and Microsoft recommends near-term expiry for ad hoc SAS tokens and notes that the account owner is billed for any usage made through a SAS, including downloads.
Seller-owned drives produce the least automatic evidence, so the custody log carries the weight: seal numbers recorded at packing, the courier tracking reference, the recipient's photo or note that the seal is intact, and the recipient's re-computed manifest hash. The key or PIN goes by a separate channel only after the seal and hashes check out. Budget the staff hours for this in the worksheet, because it repeats with every shipment.
| Control | Cloud transfer | Managed staging | Encrypted drives |
|---|---|---|---|
| Integrity evidence | Tool checksum report per job | Recipient manifest hash after pull | Manifest on drive plus recipient re-hash |
| Access expiry | Revoke grant after receipt | Link or account expiry, then delete staging copy | Key or PIN released once; drive returned or wiped |
| Custody evidence | Service logs and job IDs | Endpoint access logs | Seal numbers, tracking references, handoff log |
| Extra copies to delete | None beyond recipient | Staging copy | Drive copy, plus device wipe record |
| Main failure mode | Over-broad or unrevoked grant | Long-lived link forwarded or reused | Key shipped with drive or drive lost |
Three worked scenarios: 2 TB, 50 TB and 300 TB
Illustrative scenarios, invented to show the arithmetic. Bandwidth, utilization and drive speeds are assumptions; replace them with your own measurements. No prices are shown; the cost lines name quantities only.
Scenario A, 2 TB from a cloud bucket, 500 Mbps sustained, utilization 0.7. Network time is 2 x 8,000,000 / 350 / 3,600, about 12.7 hours. Cloud transfer or managed staging both fit a one-week window easily. Choose direct transfer if the recipient accepts a scoped grant; otherwise stage it with links that expire within the window. Cost lines: egress on 2,000 GB, a few days of staging storage, about half a day of staff time. Drives would add packing, courier and custody logging for no gain in time, so they are rejected in one line in the specification.
Scenario B, 50 TB on premises, 1,000 Mbps uplink, utilization 0.7. Network time is 50 x 8,000,000 / 700 / 3,600, about 159 hours or 6.6 days of continuous copying, before verification. At 200 Mbps with the same utilization it becomes about 794 hours, roughly 33 days. With four drives written in parallel at an assumed 200 MB per second each, copy-out takes about 17 hours, then transit and a similar copy-in. Here bandwidth decides: at 1,000 Mbps the network fits a monthly window; at 200 Mbps drives win clearly.
Scenario C, 300 TB of media files, 1,000 Mbps, utilization 0.7. Network time is about 952 hours, close to 40 days of continuous copying. A provider device service or several shipments of drives is the realistic route. Microsoft lists Azure Data Box models with 120 TB and 525 TB of usable capacity, so one larger device or several smaller ones may cover the package; check current models and regional availability first. Whichever route is chosen, the provider's manifest and erasure records belong in the transaction file alongside your own manifest hash.
Which assumption is most likely to flip the decision?
Utilization and repeat frequency flip more decisions than unit prices do. A link that tests at 1,000 Mbps but can only be used overnight has a utilization nearer one third than 0.7, which roughly doubles network time and can push a 50 TB job from about one week to about two. Show that uncertainty by running the worksheet at a low, expected and high utilization and reporting the range, not a single figure.
Repeat deliveries change the economics in the other direction. A drive shipment that wins for one 50 TB delivery can lose to a network method for a monthly 2 TB increment, because drive handling and custody work repeat every time while a configured transfer job does not.
Availability is the third assumption to check. AWS says Snowball Edge became available only to existing customers effective November 7, 2025, and points new customers to alternatives such as AWS DataSync and AWS Data Transfer Terminal. Google's Transfer Appliance and Azure Data Box remain options, with region and model limits that you confirm when ordering.
How should the chosen method be written into the delivery specification?
Once the worksheet points to a method, write the result into the delivery specification with the inputs that produced it. A recipient who can see the bandwidth, utilization and window assumptions can challenge them before the transfer instead of after a missed date.
Record the alternatives you rejected and why, in one line each. If the chosen method fails partway, for example a network transfer that runs at half the expected speed, the specification already names the fallback and the point at which you switch, which saves a renegotiation under pressure.
- Method and tool, including whether the copy is direct, staged or on drives.
- Package size from the manifest, and the manifest hash the recipient will check.
- Bandwidth and utilization assumed, with the low and high cases and resulting times.
- Who pays each cost line, especially egress and staging storage, named per party.
- Access expiry for grants, links or accounts, and who confirms closure.
- Copies to be deleted afterward, such as staging copies and drives, and the proof expected for each.
- Fallback method and the trigger for switching, for example network progress below half the planned rate after one day.
Where the worksheet stops and specialist review starts
The worksheet compares time, cost lines and evidence; it does not decide whether a dataset may leave the company or how it must be de-identified. Bring in the right reviewer when an input falls outside what arithmetic can settle.
At a glance
Where the worksheet stops and specialist review starts
- Security engineer: the recipient wants a long-lived link, a shared account or a drive without a validated encryption module. NIST's CMVP search lists validated modules and notes the user must verify the version procured is the validated version.
- Security engineer: the transfer crosses clouds or accounts that neither party fully administers, or a managed SFTP endpoint will be exposed to the internet.
- Counsel: the agreement does not say who pays egress, who owns the staging copy, or how deletion of staging copies and drives is proven.
- Privacy review: any staging copy or drive could hold personal data outside the region the agreement or your obligations allow.
- Operations: the worksheet shows network time above the delivery window at the low utilization case; plan drives or renegotiate the window before signing, not after.
How SourceX handles very large packages
For SourceX deals the large-volume rows of the worksheet matter most, because SourceX does not host multi-terabyte datasets; those packages either stay in seller-controlled storage or move on encrypted drives. Each deal's signed Dataset Deal Confirmation fixes the delivery date, a window worth running through the network-time formula above.
De-identification is done by SourceX through approved processing routes, with qualification, rights review, contracting, preparation, delivery and payment around it. Eligibility is checked through a brief company assessment that involves no upload.
Frequently asked questions
Is it cheaper to ship drives or transfer data over the internet?
It depends on volume, sustained bandwidth and how often you deliver. For a single very large package on a slow link, drives usually save time and often staff effort; for small or recurring deliveries, a configured network transfer is usually cheaper overall. Run the worksheet with your own rates rather than relying on a rule of thumb.
How long does it take to transfer 50 TB over a 1 Gbps link?
At full use of a 1,000 Mbps link, 50 decimal terabytes take about 111 hours. At a more realistic 70 percent utilization it is about 159 hours, roughly 6.6 days, before verification. Measure your real sustained throughput first.
Is managed staging less secure than direct cloud transfer?
Not inherently, but it creates one more copy and usually relies on links or accounts that can be forwarded or left open. Short expiry, one account per recipient, and deleting the staging copy after receipt close most of that gap.
Who should pay egress charges on a dataset delivery?
Whoever the agreement names. Put the payer in the delivery specification before the transfer, because cloud providers bill the storage owner by default and a shared link lets anyone holding it generate charges for that owner.
Do encrypted drives need a FIPS-validated module?
Not always by law, but a validated module gives both parties an independent check of the encryption. If you rely on validation, confirm in NIST's CMVP listing that the exact module version in the drive is the validated one.
Sources
- Microsoft's Azure Data Box overview lists devices with 120 TB and 525 TB usable capacity, AES 256-bit encryption at all times, and NIST 800-88 wiping after upload. Source
- Azure Data Box import orders produce a manifest with file sizes and checksums, a copy log during upload and chain of custody logs after data erasure. Source
- Google's Transfer Appliance specifications list customer-managed encryption keys and a unique PIN among the appliance's security features. Source
- AWS says Snowball Edge is available only to existing customers effective November 7, 2025, and lists alternatives including AWS DataSync and AWS Data Transfer Terminal. Source
- AWS DataSync checksums data during transfer and can be set to verify only transferred data or all data at the end of a run. Source
- Google's Storage Transfer Service validates checksums end to end when the source provides checksum metadata and records a failure if they do not match. Source
- Amazon S3 presigned URLs can be valid for up to 7 days when created with the CLI or SDKs and up to 12 hours from the console. Source
- Microsoft recommends near-term expiry for ad hoc SAS tokens and notes the account is billed for any usage made through a SAS, including egress. Source
- AWS Transfer Family provides managed SFTP endpoints in which each user gets a home directory in an Amazon S3 bucket, with an option restricting users to it. Source
- NIST's Cryptographic Module Validation Program lists validated modules and notes users must verify the procured version is the validated version. Source
See if your company qualifies
A short company assessment. No data uploads are needed.