BlobRouterStorage Decision Engine
Decision Report
150 TB · AWS S3 · 120 days observed · scale band 4 · list pricing
Current / month
$8,798
Recommended / month
$3,783
Monthly opportunity
$5,015
29-day break-even · 35.9x ROI at 36 months · strategy custom-per-segment
Confidence
out of 100
Do not move
compute topologyA storage-price-only model would move this data to cheaper storage. The engine prices what happens after the move.
DO NOT MOVE — Model checkpoints (80 TB stays on aws-s3)
$4,381/ mo transfer penalty if moved
- Compute reading this data lives in aws-s3 (computeDependency: same_cloud, 95% of reads from in-cloud compute)
- Same-cloud reads are modeled as free; a cross-provider move bills every read at aws-s3 egress rates
- Data criticality high — kept on aws-s3 INTELLIGENT_TIERING rather than moved
Moving 80 TB to backblaze-b2 adds ~$4,381/mo in cross-provider transfer while compute stays in aws-s3. Current placement costs $2,186/mo in storage.
DO NOT MOVE — Training data (30 TB stays on aws-s3)
$984/ mo transfer penalty if moved
- Compute reading this data lives in aws-s3 (computeDependency: same_cloud, 80% of reads from in-cloud compute)
- Same-cloud reads are modeled as free; a cross-provider move bills every read at aws-s3 egress rates
- Data criticality critical — kept on aws-s3 GLACIER_INSTANT_RETRIEVAL rather than moved
Moving 30 TB to backblaze-b2 adds ~$984/mo in cross-provider transfer while compute stays in aws-s3. Current placement costs $342/mo in storage.
Where the money goes
20% of the data produces 60% of the opportunity. Inference assets is the segment worth acting on — not the largest one.
Inference assets30 TB
$3,752 → $747−$3,005/mo
Moves to Cloudflare R2 · Standard · compute none · criticality high
Model checkpoints80 TB
$3,267 → $2,417−$850/mo
Moves to AWS S3 · Intelligent-Tiering · compute same_cloud · criticality high
Training data30 TB
$1,384 → $587−$797/mo
Moves to AWS S3 · Glacier Instant Retrieval · compute same_cloud · criticality critical
Logs10 TB
$395 → $32−$363/mo
Moves to AWS S3 · Glacier Deep Archive · compute none · criticality low
Total
$8,798 → $3,783−$5,015/mo
Year-1 average monthly cost under the Normal scenario; storage from measured segment size. Egress allocated by this segment's observed share of workload egress.
Recommended architecture
The answer is not a provider. It is a placement per segment.
80 TB · automated tiering
Model checkpoints
AWS S3 · Intelligent-Tiering
30 TB · hot edge delivery
Inference assets
Cloudflare R2 · Standard
30 TB · deep archive
Training data
AWS S3 · Glacier Instant Retrieval
10 TB · deep archive
Logs
AWS S3 · Glacier Deep Archive
Move Inference assets (30 TB) to cloudflare-r2 default
- 17 TB/mo egress (31% of workload egress) priced at cloudflare-r2 $0/GB
- No same-cloud compute dependency — moving this segment does not create a recurring transfer tax
- User-facing with strict latency requirement — placed on hot_edge_delivery
Assumptions and guardrails(expand)
Pricing
Test-only list rates (FALLBACK_PRICING). Absolute dollars are directional; rankings and topology conclusions are the output.
Measured
Modeled
Scenarios
Quoted economics use the normal scenario: year 3 at $259,394.
- Optimistic: $158,675 at year 3
- Aggressive: $489,260 at year 3
Worst case (gated): modeled 3-year cost $3,926,887 (15.1x Normal) — growth compounding is not calibrated on real bills, so this figure is not quoted as a forecast
Scenarios are labeled assumption bands (growth and egress multipliers), not statistical confidence intervals.
Horizon
Modeled over 60 months; return on investment and savings are quoted at 36 months. Observation window 120 days, data quality excellent.
Storage maturity
74/100 — descriptive only, does not affect ranking.
This is a representative AI startup workload, not a customer account. The engine runs live on the profile; dollars use published list rates and are directional. The rankings and the topology conclusions are the output.