6.2 AWS Snow Family — Snowball Edge Devices
Key Takeaways
- The Snow Family provides ruggedized physical devices for offline data transfer when network bandwidth, time, or cost make online transfer impractical.
- As of 2026 the Snow Family is two devices: Snowball Edge Storage Optimized (~210 TB usable NVMe) for bulk transfer, and Snowball Edge Compute Optimized (104 vCPUs, up to 416 GB RAM, optional GPU) for edge compute.
- AWS retired Snowmobile (the 100 PB truck) in April 2024 and discontinued Snowcone on November 12, 2024 — but the SAA-C03 blueprint still references them, so know what each was for.
- Rule of thumb: if an online transfer would take longer than about one week, evaluate Snow Family; for true edge compute with no reliable internet, choose the Compute Optimized device.
- All Snow devices use 256-bit encryption with AWS KMS keys, tamper-evident enclosures, and a TPM, and AWS performs secure erasure (NIST media sanitization) after ingest.
What the Snow Family Is For
Quick Answer: Use Snow Family for offline transfer when shipping a disk beats the wire. The current lineup is Snowball Edge Storage Optimized (bulk transfer, large NVMe capacity) and Snowball Edge Compute Optimized (104 vCPUs, GPU option, edge inference). The retired Snowmobile (100 PB truck) and Snowcone (8–14 TB portable) still appear in SAA-C03 questions, so learn their roles.
The core decision is online versus offline. Online services (DataSync, Direct Connect, S3 Transfer Acceleration) need usable bandwidth; Snow devices physically ship the data, sidestepping the network entirely. The classic exam signal is a stated slow or metered link plus terabytes of data plus a deadline.
Current Snow Family (2026)
| Device | Storage | Compute | Primary use |
|---|---|---|---|
| Snowball Edge Storage Optimized | ~210 TB usable NVMe | 104 vCPUs, up to 416 GB RAM | Large offline data migration |
| Snowball Edge Compute Optimized | 28 TB NVMe | 104 vCPUs, up to 416 GB RAM, optional GPU | Edge ML inference, video/sensor analysis at remote sites |
Retired — but still on the exam
| Device | Capacity | Status |
|---|---|---|
| Snowmobile | 100 PB (45-foot truck) | Retired April 2024 — newer services made it obsolete |
| Snowcone (HDD/SSD) | 8 TB / 14 TB | Ordering discontinued November 12, 2024 |
Why this matters: the SAA-C03 question pool (blueprint frozen August 2022) can still describe an "exabyte data-center evacuation" expecting Snowmobile, or a "small, rugged, two-pound edge device" expecting Snowcone. Answer with the device the question implies; in real architecture today you would use multiple Snowball Edge devices or online transfer instead.
Choosing Online vs. Offline, and Security
Estimate transfer time before recommending a device. A useful rule: at a given link speed, days ≈ (data in bits) / (usable bits per second) / 86,400. If the result exceeds roughly a week — or the link is needed for production traffic — go offline.
| Data volume | Link speed | Approx. online time | Recommendation |
|---|---|---|---|
| 10 TB | 100 Mbps | ~9 days | One Snowball Edge |
| 100 TB | 1 Gbps | ~12 days | One Snowball Edge Storage Optimized |
| 500 TB | 1 Gbps | ~60 days | Several Snowball Edge devices |
| 1 PB | 10 Gbps | ~12 days | Direct Connect or multiple Snowball Edge |
| 5+ PB | metered/poor | months | Many Snowball Edge (Snowmobile retired) |
Security model
| Control | Detail |
|---|---|
| Encryption | 256-bit, keys managed in AWS KMS; data is never decrypted on the device |
| Tamper resistance | Tamper-evident enclosure plus a Trusted Platform Module (TPM) |
| Tracking | E-ink shipping label and shipment tracking; chain of custody from dispatch to ingest |
| Erasure | After data lands in S3, AWS performs NIST 800-88 secure erasure of the device |
Edge compute on Snowball Edge
The Compute Optimized device runs EC2-compatible instances from AMIs you preload, AWS Lambda functions via AWS IoT Greengrass, and Amazon EKS Anywhere for Kubernetes — all while disconnected. Typical scenarios: a cargo ship classifying camera feeds, an oil rig running predictive-maintenance models, or a forward military site collecting sensor data, with results synced to AWS when connectivity returns.
Common trap: "Establish a high-bandwidth link to keep syncing 80 TB monthly" is a Direct Connect answer, not Snow Family. Snow is for one-time or periodic bulk moves, while Direct Connect is a persistent dedicated connection. Also remember Direct Connect provisioning takes weeks, so it rarely satisfies a "transfer this in the next few days" deadline.
The Snow ingest workflow, step by step
Understanding the operational flow helps you reason about exam scenarios. First you create a Snow job in the console and AWS ships the device. On arrival you connect it to your local network and use the AWS OpsHub graphical tool (or the Snowball client CLI) to unlock the device with a manifest and unlock code, then copy data onto it; OpsHub also manages EC2 instances and services on the device for edge use. You ship the device back, and AWS ingests the data into the destination S3 bucket. Throughout, data is encrypted with KMS keys you control, and the device is wiped per NIST 800-88 after ingest.
Because the unlock code and manifest are delivered separately, a stolen device cannot be read.
Why Snowmobile and Snowcone disappeared
AWS retired Snowmobile because online options (Direct Connect, DataSync) and stacking many Snowball Edge devices became cheaper and faster than chartering a 45-foot truck, and few customers ever needed a single 100 PB move. Snowcone was discontinued as the portfolio consolidated onto the more capable Snowball Edge platform. The takeaway for design questions today: for petabyte-scale moves, parallelize multiple Snowball Edge Storage Optimized devices or use a dedicated Direct Connect link; do not propose the retired truck in a real-world recommendation even though older SAA-C03 items may still name it as the intended answer.
If a question explicitly says "exabyte-scale data-center evacuation" and lists Snowmobile, that is still the keyed answer for the current exam pool; if it instead lists only modern options, choose multiple Snowball Edge devices.
A company must transfer 80 TB of data to AWS, but their internet connection would take over two weeks and is also needed for production traffic. What is the MOST efficient solution?
A remote oil platform with no reliable internet needs to run machine-learning inference on local video feeds and sync results to AWS only when a satellite link is available. Which option fits best?