The Cloud Repatriation Debate: When AWS Costs 10-100x More Than It Should
The Platform Engineering Playbook Podcast
Duration: 13 minutes Speakers: Alex and Jordan Target Audience: Senior platform engineers, SREs, DevOps engineers with 5+ years experience
📝 Read the full blog post: Comprehensive analysis with detailed cost breakdowns, decision frameworks, and real case studies from 37signals, Dropbox, and more.
🎥 Watch on YouTube
Jordan: Today we're diving into a controversial claim that's been making the rounds in platform engineering circles. An engineer sent me an article arguing that AWS charges ten to one hundred times more than it should. My first reaction was "that's ridiculous hyperbole." Then I pulled up our actual AWS bills next to Hetzner's pricing page.
Alex: Oh boy, I know where this is going. What did you find?
Jordan: An AWS C5 instance with eighty vCPUs costs between twenty-five hundred and thirty-five hundred dollars per month depending on the region. Guess what Hetzner charges for bare metal with eighty actual cores?
Alex: I'm almost afraid to ask.
Jordan: One hundred ninety dollars per month. That's an eighteen X difference. And before you say "but that's not apples to apples," I ran the benchmarks. The Hetzner box actually performs better because you're getting dedicated cores, not virtualized ones.
Alex: Wait, hold on. You're telling me we're potentially overpaying by a factor of eighteen? And this isn't some edge case?
Jordan: That's exactly what I'm telling you. And here's the kicker - eighty-six percent of CIOs are planning some form of cloud repatriation in twenty twenty-five. Up from forty-three percent just five years ago. This isn't a fringe movement anymore.
Alex: Okay, but let's be fair here. AWS isn't just selling compute. They're selling elastic scaling, global presence, managed services. There's got to be more to this story.
Jordan: You're absolutely right, and that's what makes this conversation so interesting. We're going to dig into the real economics, look at companies that have actually made the jump, and figure out when cloud makes sense versus when it's, as one engineer put it, "highway robbery."
Alex: Let's start with the poster child for cloud repatriation - thirty-seven signals. They've been very public about their AWS exit.
Jordan: Yeah, David Heinemeier Hansson has been documenting their journey in detail. Here's the headline number: they're saving two million dollars per year after leaving AWS. And they're projecting over ten million in savings over the next five years.
Alex: But didn't they have to hire a bunch of ops people to manage their own infrastructure?
Jordan: That's the surprising part - they already had eight people on their ops team when they were on AWS. They didn't have to expand the team at all. Turns out, managing AWS at scale requires just as many people as managing your own servers.
Alex: That challenges the whole "serverless means no ops" narrative we've been sold.
Jordan: Right? And they're not alone. Dropbox saved seventy-four point six million dollars over two years by moving ninety percent of their workloads off AWS.
Alex: Those are massive numbers. But I want to dig into the hidden costs that people miss when they're comparing cloud pricing. It's not just compute costs.
Jordan: Go for it. This is where it gets really interesting.
Alex: So everyone focuses on the EC2 pricing, right? But that's just the start. Egress fees are nine cents per gigabyte. Doesn't sound like much until you realize that's ninety dollars per terabyte. If you're serving video or large datasets, that adds up fast.
Jordan: And NAT gateways. Forty-five dollars a month per gateway just to route traffic between your private subnets and the internet. Plus data processing charges on top of that.
Alex: Don't get me started on cross-AZ data transfer. You're literally paying to move data between availability zones in the same region. We've seen companies spending thousands of dollars a month just on inter-AZ traffic for their Kafka clusters.
Jordan: This is what bugs me about the "you're using cloud wrong" argument. Even when you optimize everything, follow all the best practices, you're still paying five to ten times more than bare metal or VPS solutions.
Alex: It feels like gaslighting sometimes. "Oh, your bill is high because you're not using reserved instances." Okay, we switched to reserved instances. "Oh, now you need savings plans." We got savings plans. "Oh, you should be using Graviton." It never ends.
Jordan: But wait, let me play devil's advocate here. What about elastic scaling? That's the killer feature of cloud, right? You can scale up instantly for traffic spikes.
Alex: Here's the thing though - how many companies actually have unpredictable traffic spikes? Netflix does, sure. They're spending hundreds of millions annually on AWS and it makes sense for them. But most B2B SaaS companies have incredibly predictable traffic patterns.
Jordan: That's a great point. The dirty secret is that the vast majority of companies never need to scale beyond what a couple of beefy servers can handle. That article mentioned they handle millions of requests daily on a two-server setup.
Alex: Exactly. We've been sold this idea that every startup needs to be ready to scale to Facebook levels at a moment's notice. In reality, a modern server with one twenty-eight cores and a terabyte of RAM can handle insane workloads.
Jordan: Okay, but what about global presence? If you need servers in multiple regions, cloud makes that easy.
Alex: True, but providers like Hetzner and OVH have data centers across Europe and North America now. DigitalOcean has global presence. You don't need AWS to be multi-region anymore.
Jordan: Fair enough. What about managed services though? RDS, Lambda, DynamoDB - these abstract away a lot of complexity.
Alex: This is where the economics get interesting. RDS for PostgreSQL costs about five times more than running PostgreSQL on a dedicated server. The question becomes: is that convenience worth a five X premium? For a startup burning through runway, probably not. For an enterprise with money to burn, maybe.
Jordan: Let me throw another consideration into the mix - talent. I've heard the argument that it's easier to hire AWS experts than people who can manage bare metal.
Alex: I call BS on that one. First, the engineers who can properly architect and manage AWS at scale are just as rare and expensive. Second, modern tools have made server management way more accessible. The article mentions using ChatGPT and Claude to help with Linux administration. The gatekeeping around "you need twenty years of Linux experience" is outdated.
Jordan: That's a hot take, but I think you're right. I've seen junior engineers spin up and manage Kubernetes clusters on Hetzner using tools like k3s and Terraform. The automation tools have gotten really good.
Alex: Plus, let's be honest - managing a few physical servers is often simpler than managing a complex AWS setup with dozens of services, IAM policies, VPCs, and all that complexity.
Jordan: Alright, so we've established that cloud can be ten to one hundred times more expensive, and many of the supposed benefits don't apply to most companies. But there must be scenarios where cloud genuinely makes sense. Let's build a framework here.
Alex: Absolutely. Cloud makes sense when you have truly unpredictable, spiky workloads. If you're running a Super Bowl ad or launching on Product Hunt, cloud is perfect for that.
Jordan: Also for startups in the very early stage. If your AWS bill is under ten thousand a month, the operational overhead of managing your own infrastructure probably isn't worth it. Use that time to find product-market fit instead.
Alex: Agreed. I'd also add that cloud makes sense for companies that need to be in twenty-plus regions globally. At that scale, managing your own global infrastructure becomes a real challenge.
Jordan: What about regulated industries that need specific compliance certifications? Sometimes cloud providers have those certifications that would be expensive to obtain yourself.
Alex: Good point. Though interestingly, some regulated industries are moving away from cloud for data sovereignty reasons. The EU is getting strict about where data is stored and who has access to it.
Jordan: So let's flip it around. When does it make sense to consider leaving the cloud?
Alex: The magic number seems to be around ten to twenty thousand dollars per month in cloud spend. That's when the economics of managing your own infrastructure start to make sense. You can hire a part-time ops person or dedicate an engineer to it.
Jordan: And if you have predictable workloads. If your traffic patterns are consistent, you're leaving money on the table by paying for elasticity you don't need.
Alex: Also, if egress fees are killing you. If you're serving lots of video, images, or large files, the bandwidth costs alone can justify moving off cloud.
Jordan: Let's talk about the hybrid approach, because I think that's where most companies should actually land.
Alex: Yes! This isn't binary. Keep your predictable base workloads on dedicated servers or cheaper VPS providers, and use cloud for burst capacity and managed services where they actually provide value.
Jordan: We have a client doing exactly this. They moved their main application servers and databases to Hetzner, saving about eighty percent on those workloads. But they kept Lambda for event processing and CloudFront for CDN.
Alex: That's smart. Use cloud for what it's actually good at - serverless functions, CDN, maybe some ML services - but don't run your containerized web apps there if you don't need the elasticity.
Jordan: What about the migration path? Because I can already hear CTOs saying "sure, it might be cheaper, but the migration risk isn't worth it."
Alex: The key is to start small. Don't try to migrate everything at once. Start with your development and staging environments. Those don't need to be on AWS. You can probably save fifty percent of your cloud bill just by moving non-production workloads.
Jordan: Then move your most expensive, predictable workloads. Usually, that's databases and persistent services. Keep the complicated stuff on cloud while you build confidence.
Alex: Exactly. And here's a pro tip: you don't have to own physical servers. Providers like Hetzner, OVH, and even DigitalOcean offer dedicated servers that you rent monthly. No contracts, no hardware management. It's like cloud but without the insane markups.
Jordan: Let's address the elephant in the room though. There's a whole industry built around cloud - DevOps consultants, cloud architects, FinOps specialists. They have a vested interest in keeping companies on cloud.
Alex: Oh, absolutely. The article called this out directly. When your job title is "AWS Solutions Architect," you're not going to recommend moving off AWS. There's a massive economic incentive to maintain the status quo.
Jordan: And the conference circuit, the certifications, the entire ecosystem. It's a multi-billion dollar industry built on the premise that cloud is the only way forward.
Alex: But the tide is turning. We're seeing more conferences talks about cloud repatriation. The Rails community, in particular, has been leading this charge. It's becoming okay to question the cloud orthodoxy.
Jordan: So what's your prediction for the next few years? Are we going to see a mass exodus from cloud?
Alex: I don't think it'll be a mass exodus, but I think we'll see a more nuanced approach. Companies will get smarter about what belongs in cloud versus what doesn't. The days of "cloud first" as a blanket strategy are ending.
Jordan: The pendulum is swinging back toward pragmatism. Use the right tool for the job, not the most expensive tool because it's trendy.
Alex: And I think we'll see new providers emerge that offer a middle ground. Not quite as bare bones as Hetzner, not quite as expensive as AWS. Players like Fly.io and Render are already filling this niche.
Jordan: Alright, let's bring this back to our opening question. How do you justify that ten to one hundred X cost difference to your CTO or board?
Alex: You run the actual numbers. Take your AWS bill, look at what you're actually using, price out alternatives. Include migration costs, operational overhead, everything. In most cases, the math is shocking.
Jordan: And don't forget opportunity cost. That money you're saving can fund additional engineers, marketing, or just extend your runway. Two million dollars a year, like thirty-seven signals is saving, is probably twenty engineers' salaries.
Alex: The framework is pretty simple: if you have predictable workloads, you're spending over ten thousand a month, and you have at least one person who can manage infrastructure, you should seriously evaluate alternatives.
Jordan: And even if you decide to stay on cloud, going through this exercise gives you leverage. AWS has negotiated rates for companies that threaten to leave. We've seen discounts of thirty to fifty percent just by showing them a credible alternative.
Alex: That's a great point. Competition is finally coming to the cloud market, and that's good for everyone.
Jordan: So coming back to that article's provocative title - "send this to your friend who still thinks the cloud is a good idea" - I'd revise it to "send this to your friend who hasn't done the math on cloud costs."
Alex: Because cloud can be a good idea. It just shouldn't be the default idea without understanding the true economics. The fundamentals of good engineering remain constant - use the simplest, most cost-effective solution that meets your actual needs, not your hypothetical needs.