You built a multi-cloud architecture to avoid vendor lock-in. 89% of companies run multi-cloud setups thinking they’re getting the best of whichever cloud they want to use. Meanwhile, they’re bleeding tens of thousands per month on a single hidden cost that makes leaving economically impossible.
Multi-Cloud vs Hybrid Cloud (Yes, They’re Different) #
Multi-cloud = Multiple public cloud providers (AWS + Azure + GCP) in one ecosystem. The goal? Vendor lock-in avoidance, cost optimization, best-of-breed tools.
Hybrid cloud = Your crusty on-prem data center + public cloud. Usually a transition strategy for enterprises who need to keep sensitive data behind their own firewall while using cloud elasticity for everything else.
The pitch for multi-cloud sounds amazing:
- Negotiating leverage against providers
- Failover resilience (AWS US-East-1 goes down? Lmao just route to Azure)
- Risk diversification
The reality? You just traded one problem for three simultaneous flaming dumpster fires of operational complexity.
Platform Engineering Was Supposed to Save Us #
Most platform engineering teams got too excited for developer experience (DX). They built slick internal developer platforms (IDPs) with beautiful UIs, smooth CI/CD, and self-service everything.
Cool. But they treated the underlying infrastructure like a utility - an afterthought you just assume works. What happens when you do that - well, three things break: Security (becomes a late-game patch), Compliance (also a late-game patch) and Cost (a continuously leaking wound you discover in production)
The new wave is Infrastructure Platform Engineering (IPE) - treating infrastructure as a first-class product with its own roadmap, SLOs, and cost metrics. This means centralizing policies, defining governance as code from day one, and actually managing the multi-cloud chaos instead of letting it manage you.
But even with solid IPE, there’s one cost that sneaks up and absolutely wrecks budgets: data egress fees.
The Egress Fee Trap: Cloud Providers’ Hidden Profit Center #
Here’s the scam in plain English:
Data ingress (moving data INTO the cloud) = Free BUT Data egress (moving data OUT of the cloud) = Pay per GB, and the rates are designed to hurt
Gartner estimates 10-15% of your entire cloud bill is just egress fees. Not compute. Not storage. Just moving your own data around.
Where You’re Getting Pwned #
Cross-region replication: You set up disaster recovery by replicating 10TB of S3 data between two US regions. Seems responsible, right? Amazon treats that as egress from the source region. That’s $900+/month just for DR insurance.
Analytics exports: Your team runs BigQuery analytics and exports 50GB/day to cloud storage in another region or to an external BI tool. That’s $180/month in transfer fees. Every. Single. Month.
DR test surprise: You restore 15TB of archived data from cloud to on-prem to test your disaster recovery process (which, you should absolutely do). One test = $1,200 unbudgeted egress bill. Hope finance doesn’t see that one.
The exponential trap: Egress doesn’t scale linearly. One company saw compute/storage grow 3x over 8 months. Normal. But because they added media-heavy features, egress costs jumped 15x. That’s not a typo.
Isn’t this illegal? #
Someone finally noticed this was basically a protection racket. Enter the European Union’s Digital Markets Act (DMA) - Starting January 12, 2027, the DMA will ban hyperscalers from charging egress fees when businesses switch cloud providers. Between now and then, providers can only charge reduced fees equivalent to their actual costs during the switching process.
Translation: No more charging $0.09/GB to move data that costs them $0.002/GB to transfer.
Google (and others) saw this coming. So they preemptively announced egress fee waivers - which sounds great until you read the fine print. The catch: You only get free egress if you’re closing your account permanently.
You have to:
- Delete your entire GCP account
- Promise to never come back
- Complete the migration in 60 days
It’s like they’re saying: “Sure, we’ll help you pack. But only if you swear on your life you’re moving out for good.”
This is the corporate equivalent of your ex saying they’ll give your stuff back, but only if you move to a different country and change your phone number.
Are you truly free? Or just paying three cloud bills with extra steps?
How to Fight Back #
You’re not completely cooked. Here are the actual tactics:
CDNs for content delivery: Cache your static content closer to users. This alone can cut egress by 60-80% for websites/media apps. The content doesn’t need to come from your origin data center every time.
Compress everything: Before shipping data out, compress it. Standard compression shrinks data volume by 20-40% with negligible performance impact. Why aren’t you doing this already?
Private connectivity for high-volume transfers: AWS Direct Connect, Azure ExpressRoute - if you consistently push enough data to saturate even a modest 25Gbps link, the cost savings on reduced egress fees often pay for the dedicated link itself. Plus lower latency and better security. Actually worth investigating.
Governance and visibility: Use AWS Cost Explorer, Azure Cost Management, etc. to drill specifically into egress costs. Track volume, track destinations, set automated alerts for spikes. That $1,200 DR test bill? You should get notified before it ships, not after.
If you’re in the EU: Wait until 2027 and let the DMA do the heavy lifting. Between now and then, providers can only charge reduced fees equivalent to actual costs. Document everything. When the ban hits, you’ll have leverage.
The Bottom Line #
Multi-cloud architecture isn’t inherently bad. The hyperscalers designed these fee structures to be deliberately opaque. Moving data between regions? That’s one price. Moving to a different cloud? Higher price. Moving back to your data center? Even higher. Every boundary crossing is a toll booth. You built multi-cloud for freedom and flexibility. But if moving your data costs more than your quarterly cloud budget, are you actually free? Or did you just build a more expensive cage?
This article was originally published on Substack as part of the BoFOSS publication.