Pull a real AWS S3 invoice. Not the procurement slide. The bill.
The procurement slide showed storage as a single number. The bill shows five lines: storage, data transfer out, PUT/COPY/POST/LIST, GET/SELECT, and lifecycle transitions. For most enterprises in 2026, the four lines that aren’t “storage” drive the variance. Flexera’s 2026 State of the Cloud Report finds 68% of enterprises rank cloud cost optimization as their top initiative and 29% of cloud spend is wasted, the first uptick after five years of decline.
The five line items that move
Storage at rest. S3 Standard in us-east-1: $0.023/GB-month for the first 50TB, $0.022 for 50–500TB, $0.021 above. The line procurement budgets accurately because vendors quote it.
Data transfer out. Egress to the internet runs $0.09/GB at the first tier. Same-region, same-account transfers inside AWS are free. Cross-account, cross-region, or transfers outside AWS, Snowflake in a different cloud, a neocloud GPU cluster, a partner’s analytics environment, are billed per GB. For workloads where storage feeds compute outside AWS, egress is typically 2–5x the storage line.
Request operations. PUT/COPY/POST/LIST at $0.005 per 1,000. GET/SELECT at $0.0004 per 1,000, lower rate, much higher volume. 100M GETs/month = $40 on the request line alone.
Retrieval tier transitions. Standard-IA, One Zone-IA, and Glacier classes carry per-object transition fees plus retrieval fees. Intelligent-Tiering swaps these for a monitoring fee of $0.0025 per 1,000 objects/month. Procurement’s “lifecycle savings” assumption misses the retrieval side.
Minimum storage duration. Standard-IA: 30 days. Glacier Instant/Flexible: 90. Deep Archive: 180. Delete before the minimum and the bill charges the full period anyway.
Three scenarios that show the math
Run the full line-item calculation against the workload’s actual access pattern. Single-number TCO models miss the variance every time.
Scenario A, startup, 5TB stored, 3TB monthly egress
A Series A AI startup running training twice a month against a 5TB dataset on a neocloud GPU cluster outside AWS:
- Storage at rest: 5TB × $23/TB = $115/month
- Egress: 3TB × $0.09/GB = $270/month
- PUT requests (~200K): $1/month
- GET requests (~5M): $2/month
- Total: ~$388/month, egress is 70% of the bill
Scenario B, mid-market, 50TB stored, 30TB monthly egress
SaaS analytics pushing into Snowflake hosted in a different cloud:
- Storage: 50TB × $23/TB = $1,150/month
- Egress: 30TB × $0.09/GB = $2,700/month
- Requests (~50M GET, ~2M PUT): ~$30/month
- Total: ~$3,880/month, egress still ~70%
Scenario C, enterprise, 500TB stored, 300TB monthly egress
Regulated analytics across multiple clouds:
- Storage (tiered): ~$11,050/month
- Egress: 300TB at first-tier rate = ~$27,000/month (volume discounts reduce slightly)
- Requests (~500M GET, ~20M PUT): ~$300/month
- Subtotal: ~$38,350/month, egress ~70%
Cross-region replication adds another full storage line plus replication traffic. KMS, CloudTrail, and lifecycle automation add 10–20%. Total: ~$50–55K/month against a procurement-deck headline that probably showed $11K.
Where Akave’s pricing structure changes the line items?
Akave Cloud is structurally different on the four lines that drive variance. Egress isn’t a separate per-GB line. Request operations are priced predictably and don’t escalate the way S3’s request line does at scale. Minimum storage duration isn’t embedded in tiered classes, no 30/90/180-day floor. Lifecycle transitions don’t carry separate fees because the tier structure is simpler.
Storage at rest isn’t free on Akave, the comparison shape holds, but specific dollar amounts depend on the workload’s actual access pattern. Pull current numbers against your own pattern at akave.com/pricing.
When AWS S3 still fits?
S3 fits when the access pattern is internal to AWS, reads to EC2 in the same region, Athena, Redshift, all the egress lines collapse to zero and the remaining lines are competitive enough that switching costs exceed savings. S3 also fits when the team is fully AWS-native or depends on AWS-specific features (Object Lambda, Storage Lens, Macie). The math breaks for S3 when compute or analytics runs outside AWS at scale.
The procurement question list
Five lines, five questions to ask before signing:
- Egress pattern by destination. Pull six months of CloudWatch and bucket reads by where they go.
- Small-file request rate. 100M GETs/month is $40. 5B is $2,000. The handwave matters at scale.
- Retention vs. tier policy. If retention is shorter than tier minimum, every deletion pays the full minimum.
- Retrieval pattern from cold tiers. Glacier is cheap to store, expensive to retrieve, forecast volume separately.
- Cross-region replication footprint. Each replicated region adds a full storage line plus replication traffic.
Looking ahead
Variance on the four non-storage lines has gotten too big to handwave. AI training, multi-cloud, and rising egress from analytics outside AWS all amplify it. The teams ahead of their budgets model all five lines. The architecture decision is the procurement decision.
FAQ
What’s the real cost of AWS S3 beyond the headline rate?
The headline covers storage at rest. The bill includes four more lines: data transfer out, request operations, retrieval tier transitions, and minimum-storage-duration enforcement. For workloads where compute runs outside the S3 region, egress is typically 2–5x the storage line.
How is Akave’s TCO different from AWS S3?
The four non-storage lines are structured differently. No separate per-GB egress fee. Request operations priced predictably. No minimum-storage-duration floor. No separate lifecycle transition fees. Storage at rest is real and not free, pull specific numbers against your access pattern from akave.com/pricing.
When does it make sense to stay on AWS S3?
When the workload reads its data entirely inside AWS, to EC2 in the same region, Athena, Redshift, egress is zero and the math narrows. When the team is fully AWS-native and operational overhead of a second provider exceeds the savings, S3 fits. The cases where the math breaks for S3 are workloads reading at scale into compute outside AWS.
How do I model my own S3 TCO honestly?
Pull six months of CloudWatch. Bucket egress by destination. Count actual GETs and PUTs. Check retention against tier minimums. Forecast cold-tier retrieval separately. Add replication cost per replicated region. Run the math on all five lines.
Isn’t this just trading AWS lock-in for Akave lock-in?
Akave is S3-compatible, so the application layer doesn’t lock, boto3, Snowflake external volumes, and pipeline code work across both. The portability concern is data gravity, not API lock-in. Writing S3-compatible code keeps the option open in both directions.
Further Reading
- Akave pricing
- Akave product overview
- Talk to Stefaan Vervaet, Founder & CEO at Akave: stefaan@akave.com
Sources
1. AWS S3 pricing, aws.amazon.com/s3/pricing (Standard storage tiers, egress, request operations).
2. AWS S3 minimum storage duration, aws.amazon.com/s3/storage-classes (Standard-IA, One Zone-IA, Glacier Instant Retrieval, Glacier Flexible Retrieval, Glacier Deep Archive).
3. S3 Intelligent-Tiering, aws.amazon.com/s3/storage-classes/intelligent-tiering (monitoring fee, transition mechanics).
4. Flexera 2026 State of the Cloud Report, flexera.com (68% rank cloud cost optimization as top initiative; 29% of cloud spend considered wasted, first uptick after five years of decline).