We recently helped a customer migrate from a major hyperscaler to dedicated infrastructure in our datacenters. The full migration covered compute, networking, and storage—but in this article we'll focus on one piece: their offsite backup replication.
They were sending 50TB of Veeam backups to S3-compatible storage in another region. Standard disaster recovery setup. When we broke down their storage costs, the numbers made the case for moving this workload too.
Here's what we found.
The Pricing Structure
Object storage pricing has three components. The first is what gets advertised. The other two take more digging to find.
1. Storage
The per-TB rate. This is what appears on pricing pages and comparison sites. Hyperscalers typically charge $20-23/TB/month for their standard tier.
2. API Requests
Every operation has a cost. PUT, GET, LIST, DELETE—each request is metered. Hyperscaler pricing typically runs $0.005 per 1,000 write operations, $0.0004 per 1,000 read operations.
These numbers look trivial. But consider what a backup workflow actually does: Veeam running incrementals generates thousands of PUT requests per job. Listing a bucket with 100,000 objects means 100 LIST operations. Running a restore means GET requests for every object retrieved.
A typical enterprise backup workflow generates millions of API calls per month.
3. Egress
Data in is free. Data out is not. Hyperscaler egress typically runs $87-120/TB depending on provider and region.
For backup storage, egress becomes significant during restore testing. Good practice says you should verify backups regularly—actually restore data, not just check that files exist. Downloading 5TB for monthly restore verification costs $450-600/month in egress alone.
This pricing structure makes data easy to store and expensive to retrieve.
Real Numbers: 50TB Backup Workload
Here's the breakdown for this customer. Veeam backing up 200 VMs to an offsite S3 target. Daily incrementals, weekly fulls, monthly restore verification of 10% of data.
| Component | Usage | Hyperscaler (Standard Tier) |
|---|---|---|
| Storage | 50 TB | $1,150/month |
| PUT/POST/LIST Requests | ~2.3M/month | $11.50/month |
| GET Requests | ~850K/month | $0.34/month |
| Egress (restore tests) | 5 TB/month | $450/month |
| Total | $1,612/month |
Storage was 71% of the bill. Egress was 28%. API requests were negligible for this workload, but they scale fast with smaller objects or high-frequency access patterns.
Same workload on ZERO-Z3 STANDARD_IA:
| Component | Usage | ZERO-Z3 STANDARD_IA |
|---|---|---|
| Storage | 50 TB | $440/month |
| API Requests | Unlimited | $0 (included) |
| Egress (restore tests) | 5 TB/month | $27/month |
| Total | $467/month |
| Hyperscaler | ZERO-Z3 STANDARD_IA | |
|---|---|---|
| Monthly Cost | $1,612 | $467 |
| Annual Cost | $19,344 | $5,604 |
| Annual Savings | $13,740 (71%) | |
Hyperscaler example uses typical standard tier pricing ($23/TB storage, $90/TB egress, $0.005/1K writes). ZERO-Z3 STANDARD_IA: $8.80/TB storage, $5.40/TB egress, unlimited API requests included.
How Providers Compare
The major hyperscalers have similar pricing structures.
| Provider | Storage/TB/Mo | Egress/TB | Write/1K | Read/1K |
|---|---|---|---|---|
| Hyperscaler Standard (typical) | $20-23 | $87-120 | $0.005 | $0.0004 |
| Hyperscaler Express/Premium (typical) | $150-250 | $87-120 | $0.0025 | $0.0005 |
| ZERO-Z3 STANDARD_IA | $8.80 | $5.40 | $0 | $0 |
| ZERO-Z3 STANDARD (NVMe) | $70.65 | $5.40 | $0 | $0 |
Two tiers for different workload profiles: STANDARD_IA for capacity-focused workloads like backup and archive, STANDARD (NVMe) for latency-sensitive workloads like log ingestion and real-time analytics. Both include unlimited API operations and low-cost egress.
Other Providers to Consider
When evaluating alternatives, watch for conditions that affect effective pricing:
- Minimum storage periods: Some providers charge for 30–90 days minimum, regardless of actual storage duration. Delete data after a week, pay for 90 days.
- Fair use policies: "Free egress" often comes with ratio limits. Download more than you store and pricing changes.
- Minimum billing units: Some providers bill in 1TB increments. Store 1.1TB, pay for 2TB.
- Performance tradeoffs: Lower pricing may mean lower throughput or higher latency. Check benchmarks for your workload type.
The headline rate isn't always the effective rate. We built s3compare.io to make provider comparisons easier—enter your storage, egress, and request volumes to calculate costs across 25+ providers.
When This Matters
Not every workload is affected equally by the three-part pricing model.
High Impact: Backup and Archive
Backup software is API-heavy. Lots of small objects, frequent writes, periodic large restores. The 50TB example above is representative. Egress for restore testing adds meaningful cost.
High Impact: Log Storage
Observability platforms writing to object storage. Millions of small writes, frequent queries. High request volumes even with modest storage.
High Impact: Media Streaming
Store once, stream many times. Egress-heavy by design. A video library serving 10TB/month in streams pays accordingly.
Lower Impact: Write-Once Archives
Compliance data that gets written once and rarely read. Storage dominates, egress is minimal. Hyperscaler archive tiers can be cost-effective here.
The Tradeoffs
Different providers, different strengths. Here's where we fit.
ZERO-Z3
- Predictable pricing. No per-request fees. $5.40/TB egress. No fair use limits or minimum storage periods.
- European data residency (DE, FI, NL). Own infrastructure on AS215197.
- Full S3v4 API. Works with existing tools and SDKs.
- Two tiers: STANDARD_IA ($8.80/TB) for capacity, STANDARD NVMe ($70.65/TB) for single-digit millisecond latency.
- Direct support access included. One service or twenty—same access to engineers who run the infrastructure. No support tiers to unlock.
- European regions. If you need storage in São Paulo or Sydney, that's not us.
Hyperscalers
- Global presence. Dozens of regions worldwide.
- Native integration with cloud services—Lambda triggers, Athena queries, managed analytics.
- Virtually unlimited scale with no capacity planning.
- Higher storage and egress costs. API request metering.
- Enterprise support costs extra—$5,000/month minimum plus percentage of spend.
Which Fits Your Workload?
If your architecture depends on cloud-native services like event triggers or serverless functions, staying on a hyperscaler keeps things simple. If you need global edge presence, they have the footprint.
If your workload fits one of these patterns—and European regions cover your needs—the math tends to favor alternatives:
- Backup and archive: Veeam, Commvault, restic targets. High write volumes, periodic restore testing. STANDARD_IA at $8.80/TB.
- Log and analytics storage: Loki, Mimir, ElasticSearch, Splunk backends. Millions of small writes, frequent queries.
- Video streaming and static sites: Store once, serve many times. Egress-heavy by design. Multi-region SmartDNS routing included.
- Data lakes and AI/ML datasets: TensorFlow, PyTorch, Kubeflow pipelines. Large sequential reads for training jobs.
- Real-time analytics and log ingestion: This is where STANDARD (NVMe) shines. Single-digit millisecond latency within region—comparable to hyperscaler express/premium tiers at $150-250/TB, but at $70.65/TB with unlimited API calls included. High-transactional workloads where latency matters.
If predictable billing matters, per-request fees are the thing to eliminate.