Latest / The Cloud Business Podcast with Fexingo: AWS, Azure, GCP, and Enterprise Infrastructure / The Hidden Tax of Cloud Data Egress Fees
Transcript
- Lucas: Here is a number that sounds like fiction but is absolutely real in enterprise cloud accounting. If you build an application on Amazon Web Services and store all your data in the US East region, every time a user in London or Tokyo requests a file, you pay for that bandwidth. Not just once, but repeatedly. This is called egress fees, and it is effectively a tax on your own growth. Luna: Right, because the data lives there, so moving it out costs money. It feels like renting a warehouse where you get charged to take your own boxes off the premises. Lucas: Exactly that analogy works perfectly. And the rate isn't trivial. We are talking about roughly nine cents per gigabyte for standard HTTP traffic. For a small app, that is noise. But for a video streaming service or a large-scale SaaS platform serving millions of users globally, that line item can easily exceed the cost of the compute itself. Luna: So if you have ten petabytes of outbound traffic a month, we are looking at nearly a million dollars just in transfer fees. That changes the entire P&L structure. Lucas: It does. And what is interesting is that the providers frame this as a network maintenance cost. They say, 'We built the pipes, we maintain the backbone, you use them.' But from a business perspective, it creates a powerful incentive to keep data local. It is a soft form of vendor lock-in that doesn't require complex legal contracts, just cold hard math. Luna: I remember reading about a gaming studio that had to rethink their server placement entirely because their European players were triggering those cross-region charges. Did you see that breakdown? Lucas: Yes, the Blizzard expansion case was a good example. They initially tried to run everything through a central US hub to simplify management, but the latency and the egress costs made it unviable. They ended up decentralizing, which increased their operational complexity but saved them millions annually. Luna: So the trade-off is simpler operations versus lower variable costs. Which usually wins in the long run? Lucas: Usually the variable costs win, especially as you scale. But there is a second layer to this that people miss. It is not just about where the data lives, but how you deliver it. If you are serving static assets like images or videos, you should never be hitting the origin server directly. Luna: You mean using a Content Delivery Network? Because then the data is already cached closer to the user, right? Lucas: Precisely. When you use a CDN, the request is fulfilled from an edge node. The provider might still charge you for the initial pull from the origin, but that happens once. Then every subsequent view is served from the cache. For high-traffic sites, CDNs can reduce egress costs by eighty percent or more because you stop paying for repeated transfers of the same content. Luna: That seems like basic hygiene, yet I talk to CTOs who treat CDNs as an optional luxury rather than a core infrastructure component. Lucas: It is often treated that way until the bill arrives. And here is the thing about CDNs that makes the economics even more nuanced. Some providers offer unlimited egress for certain tiers. If you are running a massive operation, negotiating a custom enterprise agreement can sometimes wipe out those ninety-cent-per-gigabyte rates entirely. Luna: So it is not a fixed rule, it is a negotiation leverage point. The more volume you push, the more power you have to rewrite the pricing schedule. Lucas: Correct. And that is why large enterprises rarely look at public pricing lists. They work with account managers to get bundled discounts. But for the average startup or mid-sized tech company, the public price is the reality they face every single day. Luna: Which brings us back to the architectural decision. Where do you put your primary database if you know egress is going to be expensive? Lucas: You put it close to your largest user base. This is the concept of data gravity. Data wants to stay where the value is generated. If your customers are in Germany, your database should probably be in Frankfurt. It minimizes latency, which improves user experience, and it minimizes egress fees, which improves your margins. Luna: But what if your customer base is truly global? You can't have a database in every country. That gets messy fast. Lucas: It does get messy. That is why multi-region replication exists. You replicate data asynchronously between regions. So you have a primary in the US and a secondary in Europe. Reads can happen locally in Europe, while writes go to the US. It balances the load, but you still pay for the replication traffic between the regions. Luna: Inter-region traffic. So you are paying twice. Once to replicate, and then potentially again to serve. Is there a way to avoid that double dip? Lucas: Some providers have started offering free or discounted inter-region replication within the same continent. It is a subtle nudge to keep ecosystems regional. If you build in North America, they want your copies to stay in North America. Breaking that boundary triggers the premium rates. Luna: It feels like the cloud providers are building moats with their pricing structures rather than their technology alone. Lucas: They absolutely are. Technology is commoditized now. Everyone has good servers. Good storage. The differentiator is often the pricing model that makes it painful to leave. Egress fees are the stick that keeps you tethered to the cloud you chose. Luna: So for listeners running their own budgets, what is the first step to auditing this? Do you just look at the total bill? Lucas: No, you drill into the network section specifically. Most dashboards bury it. You want to look at outbound data transfer by destination. If you see spikes correlating with user growth in new geographies, that is your signal. You need to ask whether those users are being served efficiently or if you are burning cash on cross-ocean hops. Luna: And if you find that burn, is the fix always to move the data, or are there software-level fixes? Lucas: Software fixes matter too. Compression is huge. If you are sending uncompressed JSON responses over the wire, you are paying for redundant bits. Gzip or Brotli compression can cut payload sizes by seventy percent. That is free money if you implement it correctly. Luna: Compressing the response body. Simple, effective, and completely overlooked by developers focused on feature velocity. Lucas: Exactly. Features sell the product. Infrastructure efficiency saves the company. It is a classic tension. But ignoring egress costs is like ignoring a leak in the hull. It seems small until the ship starts sinking. Luna: Speaking of keeping things sustainable, I want to pause here for a moment. We deliberately don't run ads on these shows because we believe in supporting independent voices directly. Lucas: If you find these conversations useful for what you're building or running, consider supporting us. The link is buy me a coffee dot com slash fexingo. It helps cover the production costs and keeps the signal clean. Luna: Thanks for listening. Now, back to the numbers. Lucas: Let's talk about a specific counter-strategy. Some companies are moving away from traditional cloud object storage for archival data. They are using cheaper, specialized storage classes that have lower egress penalties, or they are accepting that retrieval will be slow and batched. Luna: Glacier or Deep Archive types. You pay less to store, but it costs more and takes longer to get the data back out. Lucas: Yes, and that trade-off is critical for compliance data. You don't need instant access to five-year-old tax records. You need them to be cheap to hold. But if you misclassify active data as archive, you will pay retrieval fees that shock your finance team. Misclassification is another hidden cost driver. Luna: So tagging and lifecycle policies aren't just organizational tools, they are financial controls. You have to automate when data moves from hot to cold storage. Lucas: Automate it or lose money. Manual processes are too slow for modern cloud environments. If you rely on humans to decide when to move data, the data stays hot too long, and you overpay. Automation aligns storage costs with actual usage patterns. Luna: It really highlights how much of the cloud bill is behavioral. It reflects how lazy or efficient our engineering teams are. Lucas: Behavioral economics applied to infrastructure. I like that framing. And it ties back to the original point about egress. It is not just a technical fee, it is a design flaw in the early internet philosophy that assumed bandwidth would be free forever. Luna: Bandwidth was never free. It just took decades for the bills to catch up with the expectations. Lucas: Precisely. The cloud providers are correcting for that historical anomaly. They are monetizing the physical reality of moving electrons across cables. And as AI models grow larger, requiring massive data movement for training and inference, those egress fees are only going to become more significant. Luna: AI workloads are data-heavy. If you train a model in one region and deploy it in another, the weight files are terabytes in size. Those transfers add up instantly. Lucas: Which is why we are seeing a rise in hybrid cloud setups for AI. Companies are keeping their heavy lifting on-premise or in dedicated facilities to avoid the cloud transfer tax for internal workflows. They only use the cloud for what it is best at: scaling user-facing services. Luna: So the cloud isn't the answer for everything. It is becoming a specialized tool for distribution, while compute stays closer to the source. Lucas: That is the trend. Specialization. Just like we specialized our applications into microservices, we are specializing our infrastructure based on cost and latency profiles. Egress fees are forcing that discipline. Luna: It forces clarity. You can no longer just throw money at a problem and hope it scales. You have to understand the physics of your data. Lucas: And that understanding is what separates mature tech companies from the ones that get surprised by their quarterly invoices. Next time you architect a new system, ask yourself where the data lives and who pays to move it. Luna: Because eventually, someone always does.