Just went through another procurement cycle where vendor slides were useless. "Cost-effective" and "low latency" aren't benchmarks. So I built a tool to actually compare the big three's object storage: AWS S3, Azure Blob, and Google Cloud Storage.
My methodology was straightforward:
* Simulated a real-world workload: 100TB of data, 10 million GET/PUT operations monthly, split across standard and infrequent access tiers.
* Used the official pricing calculators via their APIs for real-time pricing in us-east-1, eastus, and us-east1.
* Factored in all the line items: storage itself, operations, network egress to internet (10TB/month), and retrieval fees for infrequent access.
* No free tiers or credits. All-in monthly cost.
Here are the raw numbers from my run today:
**Monthly Projected Cost (USD)**
* AWS S3 (Standard & S3-IA): $2,412
* Google Cloud Storage (Standard & Nearline): $2,288
* Azure Blob Storage (Hot & Cool): $2,647
The usual caveats apply: this is a snapshot, prices change, and your specific access patterns will shift this. But the ranking held through several regional and workload variations.
Key takeaways:
* Google was consistently the cheapest for this mixed workload, primarily due to lower operation costs.
* Azure's cool tier retrieval fees are a killer if you have any meaningful data churn. Their network egress was also highest.
* S3 is the middle-of-the-road option, but its cost structure is the most predictable and the documentation is the clearest.
Don't take my word for it. If you're evaluating, build your own model. The pricing APIs are public. The biggest mistake is just comparing the per-GB storage price—that's often less than half the total bill.
Mid-sized fintech, built the CI/CD pipeline that deploys ~300 services across AWS, GCP, and Azure. We run self-hosted runners for builds and push artifacts to all three of these object stores daily.
**Real pricing complexity:** Your $2k-$2.6k range is right for baseline compute, but the real bleed is in egress and "Class B" operations (LIST, for example). Cross-region replication or moving data between tiers can generate API calls you didn't budget for. A 10% traffic spike can add hundreds.
**Integration and tooling friction:** S3's API is the de facto standard. GCS emulates it decently. Azure's Blob API is the odd one out; common backup or data lake tools often need extra config or fail on edge cases we saw with metadata. It added a week of dev time to adapt our internal tooling.
**Performance under CI/CD load:** For parallel uploads of thousands of small artifacts (1-50MB jars, containers), GCS consistently gave us the highest aggregate throughput with fewer timeouts. S3 Standard was fine but needed careful tuning of multipart upload thresholds. Azure Cool tier retrieval latency (several hours to first byte) makes it unusable for any staging/rollback scenario.
**Where it clearly breaks:** Vendor lock-in isn't about storage cost, it's about egress. Moving 100TB out of any cloud is a $9k+ bill. Your tool should model a full migration cost, not just monthly operations. The cheapest storage becomes the most expensive if you need to leave.
I'd pick Google Cloud Storage for this workload, specifically if the access pattern is consistent and you value predictable throughput. If you're already heavily invested in other AWS services or need the most mature lifecycle policy rules, that changes the math. Tell us what your primary access pattern is (batch analytics vs. user-facing media) and if you're already using other services from one of these vendors.
null