Disclaimer: I hate writing. I’m using AI to get my ideas onto paper. The opinions, experience, and numbers are mine. The grammar is not.
Let’s talk about the dirty secret of cloud computing: data transfer pricing.
AWS, Google, and Azure have all built walled gardens. The walls aren’t made of features or ecosystem lock-in. They’re made of egress bills. They’ve artificially priced network traffic to the point where moving your own data has become a tax on your business. And like any good tax, it’s one you don’t fully understand until you’re already deep in it.
The AWS Tax
On AWS, the tax is twofold.
First, there’s egress to the public internet: $0.09/GB. Then there’s cross-AZ traffic: $0.01/GB, charged in both directions. That means every cross-AZ round trip is effectively $0.02/GB. That’s the same effective rate as cross-region transfers ($0.02/GB one way). Read that again. AWS charges you the same to move data between us-east-1a and us-east-1b as it does to move data from us-east-1 to us-west-2. The only difference is how they bill the directions.
Sounds small until you realize what that actually means for how you build things.
AWS spends an enormous amount of energy promoting high availability. Multi-AZ deployments. Cross-region replication. Redundancy everywhere. It’s baked into every Well-Architected whitepaper and every Solutions Architect exam. And then they charge you every time your services talk to each other across availability zones.
It’s like a drug dealer telling you how healthy meth is. They’re pushing architectures that generate massive bills, and the customer is the one who ends up paying for it.
Let’s do the math
When I was running Rave on AWS, we were spending thousands of dollars a month on data egress alone. Not compute. Not storage. Just the privilege of moving data around.
So I started looking for alternatives. I came across OCI in 2022 and honestly thought the pricing page had a typo. North America and EU egress on OCI is $0.0085/GB. That’s not a small discount. That’s an order of magnitude.
Here’s what 200 TiB (204,800 GB) of US egress looks like on each platform:
AWS (Tiered Pricing)
| Tier | Range | Rate | Volume | Cost |
|---|---|---|---|---|
| Free | First 100 GB | $0.00/GB | 100 GB | $0.00 |
| Tier 1 | 100 GB – 10 TB | $0.09/GB | 9,900 GB | $891.00 |
| Tier 2 | 10 TB – 50 TB | $0.085/GB | 40,000 GB | $3,400.00 |
| Tier 3 | 50 TB – 150 TB | $0.07/GB | 100,000 GB | $7,000.00 |
| Tier 4 | Over 150 TB | $0.05/GB | 54,800 GB | $2,740.00 |
| Total | 204,800 GB | $14,031.00 |
OCI (Flat Pricing)
| Tier | Range | Rate | Volume | Cost |
|---|---|---|---|---|
| Free | First 10 TB | $0.00/GB | 10,000 GB | $0.00 |
| Standard | Over 10 TB | $0.0085/GB | 194,800 GB | $1,655.80 |
| Total | 204,800 GB | $1,655.80 |
The Difference
| AWS | OCI | Savings | |
|---|---|---|---|
| Monthly Cost | $14,031.00 | $1,655.80 | $12,375.20 |
| Annual Cost | $168,372.00 | $19,869.60 | $148,502.40 |
That’s an 88% reduction. On a single line item.
Let that sink in for a second. Your CFO and CEO are getting a chubby right now.
And OCI doesn’t charge for internal networking. Zero. Traffic between availability domains within a region, between subnets, between VCNs. Free. On AWS, that same internal traffic is a profit center.
”But you can negotiate an EDP”
Sure. Let’s talk about that.
On AWS, if you want reasonable egress pricing, you need to be big enough to negotiate an Enterprise Discount Program. And “commit” should be a four letter word in cloud. Because that’s exactly what it is. A commit means you have no power. You can’t leave. You have to forecast your spend years in advance. You can’t experiment freely because every dollar spent on exploration is a dollar that has to fit inside the commit. And if you optimize your infrastructure and your spend drops below the commitment? You’re penalized for being efficient. You’re a lamb in the woods, and AWS is the wolf.
What sounds better to you: getting a fair price on day one with no strings attached, or spending thousands of dollars in meetings with account managers to build a multi-year commitment that punishes you for getting better at your job?
OCI’s pricing is public. It’s the same for everyone. No private pricing agreements. No growth targets. No traps.
What free internal networking actually unlocks
This isn’t just a cost saving. It changes what you can build.
Third-party object storage without penalties. On AWS, pulling data from S3 into your compute layer across AZs costs money. On OCI, you can leverage object storage however you want without watching a meter tick.
HA systems that don’t punish you for being highly available. Anyone running HA Postgres with multi-regional replication knows exactly what this tax looks like. You see those beautiful system diagrams from companies showing off their HA Postgres setups. What they leave out is that they’re paying thousands of dollars a month just for WAL file replication across zones. That’s not a feature cost. That’s a tax on reliability.
Kubernetes clusters that bleed money from cross-talk. If you’re running a HA Kubernetes cluster, every bit of service-to-service communication across AZs is costing you. At scale, that’s thousands of dollars a month just for the privilege of having your pods talk to each other.
Load balancers that nickel-and-dime you. Running an ALB on AWS means you’re paying for egress, then you’re paying for LCUs. Arguably the least understood pricing metric in all of cloud computing. At any real scale, this adds up to thousands of dollars for the privilege of routing traffic to your application.
At the end of the day, the customer absorbs all of this. Every dollar you spend on egress is a dollar that gets passed along, one way or another.
OCI isn’t perfect
I want to be clear: OCI has plenty of sharp edges.
IAM is a dumpster fire. OCI loves massive OCIDs for everything. You want to list what’s in a bucket? Go copy the compartment ID from the console first. The compartment ID is the bane of my existence. Every CLI command, every API call, every Terraform resource needs one of these giant opaque strings. It’s maddening. Compare that to the relative simplicity of aws s3 ls and you’ll feel the pain immediately.
The dev tooling is years behind. The documentation has gaps. I’ll write about all of this too. But from a networking and data transfer standpoint, nobody can touch them right now. It’s not even close.
The bottom line
The hyperscalers have turned networking into a profit center at the expense of good architecture. They promote patterns that generate egress bills, then charge you for following their own advice. It’s a walled garden held together by data transfer fees, and most companies don’t realize how much they’re paying until they actually break it down.
If you’re spending real money on cloud infrastructure and you haven’t looked at your egress line items recently, go do that. Then come back and tell me you’re not angry.
Rant over.