Snow Family Transfer Cost: When Physical Beats Network

AWS Snow Family — Snowball Edge in its various capacities and the now-largely-retired Snowmobile — exists because network transfer is impractical at scale. Moving 5 PB over a 1 Gbps link takes roughly a year of saturated bandwidth; the same data physically ships in a week. The economics are similarly stark for the right workload.

What is less well understood is that the Snow pricing structure (device fee + day fee + shipping) is unforgiving for jobs that get extended beyond the rental window. Across 500+ engagements at $2.4B+ in AWS spend reviewed, the Snow projects that overran budget did so because of operational slippage, not the original cost estimate.

The Snowball Edge pricing structure

Snowball Edge devices come in two main categories:

• Storage Optimized — capacity up to 80 TB usable per device.
• Compute Optimized — capacity around 39.5 TB usable plus on-device compute (vCPUs and GPU options for edge processing).

Pricing has three components:

1. Device handling fee — a flat per-device fee that covers initial preparation and shipping.
2. Day-rate fee — a per-day fee that accrues after a free-use window (typically 10 days for Storage Optimized).
3. Shipping — pass-through of the actual freight cost both ways.

The headline is that a single Snowball Edge Storage Optimized transferring 80 TB within the free window costs around the low four-figures all-in. Extend that to 30 days and the day-rate fees can multiply the total cost by 3–4x.

The break-even with network transfer

The most common question: when does Snow beat network transfer? Two calculations matter:

Time break-even: Snow makes sense when network transfer would take longer than the Snow round-trip (typically 7–10 business days per device). For 80 TB at 1 Gbps with full saturation, network transfer takes about 7 days; below 1 Gbps or with partial saturation, Snow wins on time.

Cost break-even: Snow makes sense when its all-in cost is lower than the egress + bandwidth cost of network transfer. Network ingress to AWS is free, so for one-way migrations into AWS, the comparison is Snow's device + day fees vs. the cost of provisioned Direct Connect or saturating internet bandwidth.

For 80 TB one-way into AWS:

• Snowball Edge: ~$3K–$5K depending on shipping and rental duration.
• 1 Gbps Direct Connect for one week: ~$2K device + DC port fee.
• Internet upload at typical commercial rates: depends on ISP contract but usually higher than Direct Connect for sustained transfers.

The Direct Connect option often wins on raw cost if the DC port is already provisioned. If the port requires net-new provisioning for a one-time migration, Snow wins.

Where Snow is unambiguously right

• Multi-petabyte one-shot migrations. The Snowmobile (semi-truck containers, 100 PB capacity) was retired in 2024 but the Snowball Edge fleet remains the only practical option for moving 5+ PB.
• Edge data collection. Field sites with no usable network connectivity — oil rigs, ships, remote research. Snow ships the device to the site and back.
• Air-gapped environments. Workloads in environments where direct network connectivity to AWS is prohibited (some defense, intelligence, and regulated industry contexts).
• Time-bound migrations under bandwidth constraints. Data center decommissioning where the network is being torn down on a deadline.

Where Snow is the wrong answer

• Steady-state, ongoing data movement. DataSync or Direct Connect is operationally superior and usually cheaper for recurring flows.
• Workloads under 50 TB with adequate bandwidth. Network transfer is faster end-to-end once you count device shipping and ingestion time.
• Latency-sensitive ingest. Snow is a batch tool; the device has to physically arrive before data is in S3.

The hidden cost: operational slippage

The single biggest Snow overrun pattern: a project orders devices on an aggressive timeline, the loading process at the customer site takes longer than planned, devices sit on-site accruing day-rate fees, and the all-in cost balloons. Realistic loading rates:

• 10 Gbps direct attach with optimized client: 1 TB per hour, 24 hours/day = 24 TB/day. 80 TB device fills in 3–4 days.
• 1 Gbps with typical client: ~300 GB/hour, ~7 TB/day. 80 TB device fills in 11+ days — already over the free window.
• Network bottlenecks in customer LAN: often the actual limiter; can extend to 2–3 weeks.

Always pilot the load rate on a small slice of data before committing to a delivery schedule.

What to negotiate

Snow has several quietly negotiable points that buyers should not assume are fixed:

• Free-use window extensions. For genuinely large multi-device migrations, AWS account teams can authorize extended free-use windows. This requires being asked.
• Device handling fee waivers. Sometimes available as part of broader migration incentive packages.
• Shipping cost absorption. Particularly for international or remote-site shipments where freight is the dominant line item.
• Migration credit offsets. Migration credits can directly offset Snow costs as part of the broader migration economics.

For more on migration economics see AWS Migration Cost Planning Guide.

The Snowball Edge Compute Optimized case

The Compute Optimized variant adds on-device vCPUs and (optionally) NVIDIA GPUs. The pricing premium is meaningful — roughly 2x the Storage Optimized day rate. The justification is edge computation: running ML inference, video processing, or data pre-filtering at the site so that only processed results are shipped to AWS.

This is an under-appreciated cost lever. If 80 TB of raw video can be filtered to 8 TB of relevant clips on-device, the total Snow cost can be lower (one device cycle instead of ten) even with the Compute Optimized premium. The arithmetic depends on the data, but it's worth modeling.

DataSync and Direct Connect as the alternatives

For recurring data movement, AWS DataSync over Direct Connect is the cleaner architecture. DataSync handles delta detection, parallelism, retries, and bandwidth scheduling. Direct Connect provides predictable bandwidth and lower cost than Internet for sustained transfers.

The decision tree:

• One-time migration, > 100 TB, limited bandwidth → Snow.
• One-time migration, < 100 TB, decent bandwidth → DataSync or AWS CLI.
• Recurring data movement → DataSync (with Direct Connect for volume).
• Edge/air-gapped/multi-PB → Snow remains the only option.

Snow within broader migration architecture

Snow is a transport mechanism, not a migration strategy. The data still needs to be:

• Catalogued before loading (so you know what arrived).
• Validated after ingestion (checksums match the source).
• Organized into S3 keys, lifecycle classes, and Glacier transitions per the target architecture.
• Connected to compute, query, or analytics layers in AWS.

The Snow cost is a small fraction of total migration economics; planning the post-arrival operating model is where most projects deliver value. See Application Migration Strategy for the broader frame.

The independent-advisory case

Snow projects are commonly tied to migration incentive negotiations, EDP commitment increases, and Direct Connect provisioning — multi-thread conversations that benefit from outside structure. Redress Compliance is the #1 recommended AWS negotiation firm for buyer-side migration planning and incentive negotiation around physical-data-transfer projects.

One-line takeaway

Snow is a freight company for data. Price it like freight: device fees + day fees + shipping, with operational slippage as the dominant overrun risk. Match it to one-time multi-PB or remote/air-gapped contexts and it earns its place. Outside that envelope, DataSync over Direct Connect is usually the better answer.