Latest / The Cloud Business Podcast with Fexingo: AWS, Azure, GCP, and Enterprise Infrastructure / Why Cloud Bills Now Charge for Egress to On-Prem
Transcript
- Lucas: So there's a line item on enterprise cloud bills that's been around for a decade, but it's showing up with a lot more zeros lately. It's the charge for moving data out of the cloud back to on-premises. Luna: You mean egress fees. The ones everyone jokes are the real cloud tax. Lucas: Exactly. And it's not a joke when you see a forty-thousand-dollar monthly bill just for sending data to your own data center. I talked to a fintech infrastructure lead last week who said their egress cost doubled year-over-year even though their overall cloud spend stayed flat. Luna: How is that possible? Did they suddenly start moving more data? Lucas: Part of it is that hybrid architectures are back in fashion. Companies realized they can't put everything in the cloud, especially with AI training on-prem and compliance requirements. So they're syncing databases, replicating logs, moving model artifacts — and every one of those flows crosses the egress meter. Luna: And the meter runs at different rates depending on which provider and which region you're leaving. Right? Lucas: Right. AWS charges anywhere from nine cents to fifteen cents per gigabyte for the first ten terabytes a month, then it tiers down. Azure has similar tiers, and GCP is a bit cheaper in some regions but still not free. Multiply that by petabytes and you're in the six-figure territory fast. Luna: And the funny thing is, you can't just say, oh, I'll keep everything in one region. Because your on-prem data center is where it is. Lucas: Exactly. So the real question is, what can you actually do about it? And that's what we're going to dig into today. Lucas: First, let's talk about why egress is so expensive in the first place. The providers argue it's not just bandwidth — it's the cost of running the network backbone, the peering agreements, the hardware amortization. But analysts have pointed out that egress margins are a big profit driver for them. Luna: So it's a profit center disguised as a pass-through cost. Lucas: That's the perception. And it's why the EU's Digital Markets Act has been pushing for more transparency and lower rates. In 2024, the big three all cut egress prices in Europe after pressure from regulators. Luna: But not everywhere else, right? Lucas: Right. In the US and Asia, the rates stayed pretty much the same. So if you're a global company, your bill depends a lot on where your data is leaving from. Luna: That's a nice geopolitical angle. But let's get practical. What are the levers an operator actually has? Lucas: The first lever is content delivery networks. If you're serving files to users, you put a CDN in front and the traffic goes to edge locations, not back to on-prem. That can cut egress by eighty percent in some cases. Luna: But what if it's internal data, like database replication or log shipping? Lucas: That's where you start re-architecting. Some teams compress data before it leaves, which reduces the volume. Others use private connectivity like Direct Connect or ExpressRoute, which can have flat-rate pricing instead of per-gigabyte. Luna: So you're trading a variable cost for a fixed one. That's the classic move. Lucas: And it works if you move enough data. The break-even is usually around ten to twenty terabytes a month. Below that, you're better off paying per gigabyte. Luna: Good rule of thumb. But what about the AI angle? I've heard that training models on-prem and then syncing checkpoints to the cloud can be a huge egress driver. Lucas: That's a real one. A lot of companies are doing hybrid AI training — they use on-prem GPU clusters for data privacy reasons, but they want to back up checkpoints to object storage in the cloud. Those checkpoints can be tens of gigabytes each, and they sync every few minutes. Luna: That adds up fast. And it's not like you can skip the backup. Lucas: Right. So some teams are getting creative. They use incremental syncs, they only upload the final model, or they schedule the syncs during off-peak hours to take advantage of lower tier rates. But honestly, it's still a significant line item. Luna: And the providers know that. They're not going to make it easy to leave. Lucas: That's the thing. Once you have a lot of data in a cloud region, the cost to move it out is a huge switching cost. It's lock-in by design. But there's actually a movement to change that. The Open Compute Project has been working on a framework for free egress between clouds, and some smaller providers are offering zero egress as a differentiator. Luna: So it's not impossible. It just takes pressure from customers and regulators. Lucas: Exactly. And on that note, I want to mention something that keeps this show going. If today's breakdown helped you see your cloud bill a little differently, that's exactly what we're here for. And if it was worth a coffee to you, you can support us at buy me a coffee dot com slash fexingo. Luna: It does help us keep the lights on and the episodes ad-free. Appreciate anyone who's ever tossed a few bucks our way. Lucas: Alright, so back to the tactics. Another thing we're seeing is teams using data lifecycle policies to automatically tier data to cold storage before it ever reaches the egress meter. If you only need the data for compliance, you don't need to move it every day. Luna: That's smart. It's not about avoiding egress, it's about reducing the volume that needs to egress in the first place. Lucas: Right. And there's also the negotiation angle. If you're a big enough customer, you can ask for egress discounts as part of your enterprise agreement. Some teams have gotten ten to twenty percent off by threatening to move to a competitor. Luna: But that only works if you have leverage. What about the rest of us? Lucas: For smaller teams, the best move is to monitor egress with tools like Vantage or CloudHealth. You can set up alerts when egress spikes, and then you know exactly which workload is driving the cost. A lot of times it's a forgotten batch job that's pulling huge datasets every night. Luna: So it's a visibility problem first, then an optimization problem. Lucas: Exactly. And the last thing I'll say is this – egress fees aren't going away, but they are becoming more transparent. The next time you see a cloud bill, look for the egress line item. If you don't know what's generating it, that's a red flag. Luna: And that seems like a good place to end. But I'm curious – what's the one thing you'd want a listener to remember from this episode? Lucas: That egress is the one cost you can't optimize away by just shutting down instances. It's tied to your data flows, and the only way to control it is to understand your architecture and negotiate early. Don't wait until the bill arrives. Luna: Well said. Thanks for listening, everyone.