Latest / The Cloud Business Podcast with Fexingo: AWS, Azure, GCP, and Enterprise Infrastructure / Why Cloud Exit Is Actually Possible Now
Transcript
- Lucas: So there's this belief in enterprise IT that the cloud is essentially a one-way door. You move in, you pay the egress fees, you rebuild your architecture around managed services, and after a certain point, leaving would cost more than staying — so you just don't think about it. Luna: Right, the 'cloud repatriation is a myth' line that gets repeated at every AWS re:Invent keynote. Lucas: Exactly. But a handful of companies have actually done it. And the most documented case is 37signals — the company behind Basecamp and HEY. In 2022 they announced they were leaving AWS, and by early 2025 they had completed the move. Their projected savings: seven million dollars over five years. Luna: Seven million. That's not a rounding error. That's a real number that makes you ask: who else could do this? Lucas: So let's walk through exactly what they did, what it cost them to leave, and what the economics look like for a typical enterprise. Because the answer to 'can you exit the cloud?' is very different depending on who you are. Luna: Start with 37signals. What was their cloud bill before they left? Lucas: They were spending roughly three point two million dollars a year on AWS — mostly on EC2 compute, some RDS for PostgreSQL, a chunk of data transfer, and a long tail of smaller services. David Heinemeier Hansson, their CTO, went public with the breakdown. The big insight was that their workloads were stable and predictable — they weren't spinning up thousands of instances for a flash sale. They run Basecamp and HEY, which have steady usage curves. Luna: So their architecture was suited for fixed-capacity hardware. That's already different from a lot of enterprises that need elasticity. Lucas: Exactly. They bought their own servers — Dell PowerEdge, off the shelf — and colocated them. They signed a five-year hardware contract. The upfront was significant, but when you amortize it, the monthly cost is roughly forty percent less than what they were paying AWS for equivalent compute. Luna: And the data transfer? That's the hidden tax everyone talks about. Lucas: They eliminated it almost entirely. In the cloud, every gigabyte that leaves AWS to the internet costs between nine and fifteen cents. When your servers are in a colo facility, you pay a flat rate for your port — usually a few thousand dollars a month for a 10-gigabit pipe. If you're moving a lot of data, the savings are enormous. 37signals estimated they cut their bandwidth cost by about eighty percent. Luna: But there's a catch, right? They had to manage their own hardware. That means staff, monitoring, spare parts, dealing with failures. Lucas: They already had a strong ops team. That's the second big filter. If you don't have people who know how to rack and stack, or you've let those skills atrophy, the labor cost of going back on-prem is real. 37signals kept a small team — maybe two or three people focused on infrastructure — but they also automated a lot with Chef and Kubernetes. They didn't go back to manual server management. Luna: So what about the managed services tax? They were using RDS for PostgreSQL. Did they switch to self-managed Postgres? Lucas: They did. And they said it was the scariest part. RDS handles backups, failover, patching — you pay a premium for that convenience. They moved to their own PostgreSQL setup with Patroni for high availability. It required more DBA expertise, but the cost dropped significantly. They estimated RDS was costing them about two and a half times what the raw compute would have cost for equivalent database performance. Luna: That's a big premium, but for a lot of companies, that convenience is worth it. If you don't have a dedicated DBA, RDS is cheaper than hiring one. Lucas: Exactly. So the cloud exit math isn't just about hardware costs — it's about organizational capability. 37signals is a software company with a strong engineering culture. They could absorb that complexity. A mid-size retailer that moved to the cloud because they didn't want to manage servers? The calculus is different. Luna: Let's talk about the broader trend. Are we seeing more companies follow 37signals? Lucas: A few high-profile cases. Dropbox did a big repatriation in 2016 and claimed they saved seventy-five million dollars over two years. But they were a special case — they had massive storage costs and built their own infrastructure. More recently, companies like Zillow and some financial services firms have pulled certain workloads back. But it's not a flood. Most enterprises are still moving the other direction. Luna: What about the pure cloud providers? There's a whole ecosystem of bare-metal cloud services now — companies like Equinix Metal, Vultr, Hetzner. You get dedicated hardware but with cloud-like APIs and billing. Lucas: That's actually the interesting middle ground. You can avoid the managed services markup and the egress fees of the big three, while still getting automation and pay as you go pricing. Hetzner, based in Germany, is notorious for being a fraction of AWS cost for raw compute. But you don't get the ecosystem — no S3, no Lambda, no RDS. Luna: So the decision really comes down to: how much are you paying for convenience, and do you have the team to replace it? Lucas: That's the framework. And it's worth naming something here — a handful of listeners chip in monthly to keep this show ad-free and independent, and that support lets us dig into questions like this without worrying about sponsorship constraints. If these conversations are useful for what you're building or running, there's a link at buy me a coffee dot com slash fexingo. It's a small group already, and it's what keeps this going. Luna: Yeah, and it means we can spend time on the details that matter. So back to the economics — what does this mean for AWS? Lucas: AWS is a hundred billion dollar business. Even if a few percent of workloads repatriate, it's a rounding error for them. But the interesting signal is that AWS is actually responding. They've dropped egress fees for customers who want to leave — they now waive data transfer costs if you're migrating to another provider. That was unthinkable five years ago. And they've introduced savings plans and committed-use discounts that make the pricing more competitive with bare metal. Luna: So cloud exit becomes a bargaining chip. Even if you never leave, the threat of leaving can get you a better deal. Lucas: Exactly. And that's the practical takeaway. Every enterprise should run the numbers on what their workload would cost on colocated hardware or a bare-metal provider. Even if you stay, you'll negotiate from a position of knowledge. The 37signals case is useful because it's public and detailed. But most companies don't have the same profile. The real question is: do you have stable, predictable workloads, and do you have the engineering team to run infrastructure? Luna: If the answer to both is yes, the savings can be real. If not, the cloud is probably still cheaper when you include the hidden cost of labor. Lucas: Right. And that's the honest answer. Cloud exit is possible, but it's not for everyone. The smartest move might be to build a hybrid strategy — keep your elastic workloads in the cloud, move your steady-state workloads to dedicated hardware. The all-or-nothing thinking is the real trap. Luna: So the headline isn't 'cloud is dead.' It's 'know your costs, know your team, and don't assume the default is the cheapest.' Lucas: Exactly. And that's the conversation we should be having — not whether to go cloud or on-prem, but how to make the trade-off explicit. Thanks for listening.