---
title: "Why Your Egress Fees Are Quietly Killing Your Margins — And How $0 Egress on Bare Metal Changes Everything"
url: "https://bitrefinery.com/blog/egress-fees-killing-margins-zero-egress-bare-metal"
description: "Cloud egress fees are one of the most overlooked budget killers in data infrastructure. Here's what they're actually costing you — and why bare metal with $0 egress changes the math entirely."
author: "Bit Refinery Infrastructure Team"
date: "2026-04-26"
lastmod: "2026-04-26"
tags: ["bare metal", "egress fees", "cloud costs", "infrastructure", "clickhouse", "cost optimization", "data engineering"]
source: "blog CMS"
---

# Why Your Egress Fees Are Quietly Killing Your Margins — And How $0 Egress on Bare Metal Changes Everything

There's a line item on your AWS bill that probably doesn't get enough attention in your quarterly reviews. It sits there, quietly compounding, while your team debates compute costs and reserved instance strategies. It's egress. And for data-heavy workloads, it's not a rounding error — it's a margin killer.

Let's talk about what's actually happening here, because I think a lot of engineering teams don't fully internalize the math until they're staring at a $20k+ monthly bill and wondering where it all went.

## The Egress Tax Is Real, and It's Steep

AWS charges between $0.05 and $0.09 per GB for data transferred out of their network. Azure and GCP are in roughly the same ballpark. Sounds small, right? Until you run the numbers.

If you're moving 200 TB of data per month — not unusual for a ClickHouse cluster serving a busy analytics platform, or a media pipeline pushing video assets — you're looking at somewhere between **$10,000 and $18,000 a month** just in egress. Every month. That's not your compute. That's not your storage. That's just the privilege of getting your own data out. This is a significant part of what many call [the $200 billion AI tax on your cloud bill](/blog/200-billion-ai-tax-cloud-bill), where infrastructure costs balloon far beyond the initial estimates.

And it compounds in ways that catch teams off guard. Your BI dashboards query data. Your ML pipelines pull training sets. Your microservices talk to each other across availability zones. Each of those data movements has a cost attached to it, and the cloud providers have structured their pricing so that the more you use their platform, the more you pay to leave it.

This is by design. Egress fees are essentially a switching cost baked into your monthly invoice.

## Who Gets Hit Hardest

Not every workload feels this pain equally. If you're running a small SaaS app with modest data transfer, egress probably isn't your biggest problem. But if you're in any of these situations, you should be paying close attention:

**Ad tech and real-time analytics.** You're ingesting billions of events and querying them constantly. The data never stops moving.

**Healthcare and life sciences.** Medical imaging files are massive. DICOM datasets, genomic data, research pipelines — the transfer volumes add up fast and the regulatory requirements mean you can't just compress everything into oblivion.

**Video and media.** Obvious one. If you're storing and serving video, egress fees are a core part of your unit economics whether you like it or not.

**Financial services.** High-frequency data, market feeds, risk calculations running against large datasets. The latency requirements alone push you toward architectures that move a lot of data. For these teams, finding ways to [query 40+ data sources without the cloud tax](/blog/trino-on-bare-metal-federated-sql-zero-egress) becomes a competitive necessity rather than just a cost-saving measure.

For all of these, egress isn't a footnote — it's a fundamental part of the cost model.

## The Bare Metal Difference

Here's where the math gets interesting. On bare metal infrastructure with flat-rate, unlimited bandwidth — like what we offer at Bit Refinery — egress fees simply don't exist. Zero. You get a 1 Gbps connection included in your monthly price, and you can push as much data through it as you want.

Let's put that in concrete terms. Our Gold tier server — 80 cores, 1 TB RAM, 44 TB of RAID6 SSD storage — runs **$2,800/month**. A comparable AWS configuration (r6i.metal with similar storage) comes in around **$10,658/month** before you touch egress. Add 200 TB of monthly data transfer and you're looking at another $16,000+ on top of that.

That's not a 2x difference. That's a 10x difference. For the same workload.


![Cost comparison chart showing AWS vs Bit Refinery bare metal with 200TB egress fees](/api/storage/files/blog-images/infographic-1777201280310.jpg)

Now, I'm not saying bare metal is the right answer for everything. It's not. Spiky, unpredictable workloads where you need to scale from zero to a hundred in minutes — cloud makes sense there. But for baseline, predictable, data-intensive workloads? The economics of bare metal are hard to argue with.

## The Google Cloud Interconnect Angle

One thing worth mentioning specifically for teams running hybrid architectures: at our Denver facility, every server comes with a free private peering connection to Google Cloud. No port fees, no cross-connect charges, no monthly interconnect bill.

What that means in practice is that data flowing from your bare metal servers into GCP costs nothing on either side. And data coming back from GCP to your bare metal runs at roughly $0.02/GB — compared to $0.08–$0.23/GB for standard internet egress from Google Cloud. We've written extensively about [why we give away $1,500/mo in cloud interconnect fees](/blog/why-we-give-away-cloud-interconnect-fees) and how it fundamentally changes the way you can architect your data pipelines.

For teams doing GPU training on-prem and serving models through Vertex AI, or running ClickHouse on bare metal and querying BigQuery for supplemental data, this is genuinely significant. AWS Direct Connect runs $1,500–$2,250/month. Azure ExpressRoute is $5,000–$8,000/month. GCP Partner Interconnect is $1,700+/month. Ours is included.

## "But We're Already Deep in AWS"

Fair. Migration isn't free, and I'm not going to pretend it is. Rearchitecting data pipelines, retraining your team, dealing with the operational complexity of a hybrid setup — that's real work with real costs.

But here's the thing: you don't have to move everything. The [hybrid cloud model of baselining on bare metal](/blog/hybrid-cloud-bare-metal-baseline-burst-public-cloud) means you can put your steady-state, high-volume workloads on bare metal — the stuff that runs 24/7 and moves predictable amounts of data — and keep cloud for the bursty stuff. You get the cost predictability where it matters most without giving up the flexibility you actually need.

A lot of our customers started by migrating just their analytics infrastructure. Move the ClickHouse cluster to bare metal, point the dashboards at it, and watch the egress line on the AWS bill disappear. That one change often pays for the entire bare metal deployment with room to spare.

## The Bigger Picture

Egress fees are a symptom of a broader dynamic: cloud providers have structured their pricing to make it cheap to get in and expensive to get out. That's not a conspiracy theory, it's just good business on their part. But it means the total cost of ownership math looks very different at month 12 or month 24 than it did when you first spun up that cluster.

If you're running data-intensive workloads and you haven't done a real TCO analysis recently — one that includes egress, data transfer between services, and the actual cost of the compute you're using — it's worth doing. The numbers might surprise you, especially since [GCP just doubled egress rates](/blog/gcp-doubled-egress-rates-free-private-peering-hybrid-architecture) for many common scenarios.

And if you want to talk through what that looks like for your specific architecture, we're happy to dig into it. No sales pitch, just math.

[Get in touch with the Bit Refinery team →](https://bitrefinery.com/contact)