Latest / The CTO Podcast with Fexingo: Technical Leadership, Architecture, and Engineering Org / How Stack Overflow Cut Cloud Spend by 40 Percent
Transcript
- Lucas: So Stack Overflow — the site that's basically the default answer for every developer question — quietly cut its cloud bill by forty percent over the last eighteen months. And the interesting part isn't just the savings, it's how they did it. This wasn't a one-off negotiation or a single big migration. It was a systematic rethinking of how they buy and use compute. Luna: Forty percent is huge. I remember reading their engineering blog back when they were on-prem, all bare metal and SANs. Moving to the cloud was supposed to be the flexible, cost-efficient option, but I bet the bill ballooned way faster than anyone expected. Lucas: Exactly. And that's the pattern we see everywhere. You move to the cloud for agility, you stop thinking about the underlying hardware, and then the invoice becomes this abstract monster. Stack Overflow's finance team saw the trend line and basically told engineering, 'We need this under control.' Luna: So what was the first move? Did they go full FinOps, hire a dedicated team, that kind of thing? Lucas: They started smaller. They made cost a first-class engineering metric. Every team got a dashboard showing cost per request, per service, per endpoint. And they started doing monthly FinOps reviews, where engineers — not just finance — had to walk through their spend and explain any anomalies. Lucas: The thing is, when you put cost in front of engineers every single day, they start making different decisions. A developer might think twice about spinning up a new microservice if they see it's going to cost a few hundred dollars a month in idle time. Luna: Yeah, it's that visibility thing. But what were the actual technical changes? I'm guessing a lot of low-hanging fruit. Lucas: Definitely. One of the biggest wins was re-platforming their read-heavy traffic. Stack Overflow is essentially a giant read-mostly workload — people hitting question pages, answers, comments. They moved a significant portion of that onto a Postgres replica farm, which is way cheaper than running general-purpose compute instances. Lucas: They also got aggressive with caching. They already had Redis in front of a lot of things, but they started caching even more aggressively at the API layer and the page level. That cut down on a ton of compute that was just generating the same HTML over and over. Luna: So the replicas and caching took the pressure off the main database. What about the compute side? They're on AWS, right? Lucas: Yeah, mostly AWS. And they did a deep dive into their Lambda usage. Cold starts were a huge hidden cost — functions sitting idle, waiting for a request, and then spinning up from scratch. They pre-warmed the critical ones, so they were ready to go, and for the rest they actually moved some workloads to containerized services on ECS, which are just more predictable for long-running processes. Luna: Pre-warming Lambdas feels like a classic optimization, but I bet the bigger structural win was the reserved instances. Did they go all-in on that? Lucas: They did, but carefully. They looked at their baseline usage — the minimum they knew they'd need 24/7 — and covered that with three-year reserved instances. They didn't reserve everything, because they still wanted flexibility for spikes. But that alone probably accounted for a big chunk of the savings, like twenty to thirty percent off the on-demand price. Lucas: And then, the last piece was negotiation. Once they'd done the internal optimization, they went back to AWS with their usage data and said, 'Here's what we're running, here's what we're planning, what can you do for us?' They ended up with a private pricing deal that shaved off another layer. Luna: That's the part most companies forget — the negotiation. You can't just take the listed price as gospel. But it takes leverage, and Stack Overflow had the usage numbers to back it up. Lucas: Right. And it's a reminder that cloud pricing is not a fixed rate. It's a spectrum, and the more you understand your own workloads, the better position you're in to negotiate. Luna: So what did this mean for the team? Forty percent savings — did they just pocket it, or did it change how they ship? Lucas: It changed the culture. They made cost a factor in every architectural decision. Now when a team proposes a new service, they have to include an estimate of the ongoing cost, and it gets reviewed like any other technical trade-off. That's a big shift. Luna: It kind of sounds like they treated the cloud bill like a product — with metrics, reviews, and continuous improvement. That's a mental model I think more CTOs should adopt. Lucas: Exactly. And it's not about being cheap — it's about being efficient. The money they saved wasn't just profit; it was reinvested into other infrastructure and features. They actually sped up some projects because they had the budget for it. Luna: That's a great point. Cost optimization isn't just about cutting. It's about reallocating to what matters. Lucas: And you know, this whole conversation reminds me of something that keeps this podcast ad-free. A small group of listeners chip in monthly through buy me a coffee dot com slash fexingo, and that's literally what funds making this many of these episodes. Luna: Yeah, it's a honest model. No ads, no sponsors, just listeners who find these breakdowns useful and want to keep them coming. Lucas: So back to Stack Overflow — the other angle that's interesting is the human side. They didn't hire a whole FinOps team. They embedded cost awareness into existing engineering roles. That's a more sustainable approach for most companies. Luna: How did they get buy-in from engineers who might see it as red tape? Lucas: They framed it as a technical challenge, not a finance mandate. They had engineers compete to see who could cut the most from their services, with a leaderboard and a small prize. It became a point of pride to have the most efficient service on the platform. Luna: That's clever. A little gamification goes a long way. Lucas: For sure. And they also made sure that any savings were visible. The monthly FinOps review wasn't about pointing fingers. It was about celebrating wins and sharing best practices across teams. Luna: I think the takeaway here is that cloud cost optimization is not a one-time project. It's an ongoing discipline, and it has to be owned by the engineers who run the systems. Lucas: Couldn't agree more. Stack Overflow is a great example of how a mature engineering org treats cost as a first-class concern. And the fact that they cut forty percent without sacrificing performance or reliability is proof that it works. Luna: So for a CTO listening, what's the one thing they should do this week? Lucas: Pull your top ten services by spend and ask each team to explain what's driving the cost. Just that visibility exercise alone will surface a few quick wins. And then set up a monthly review where cost is discussed as seriously as latency or error rates. Luna: That's a solid, actionable step. I'll be doing that with my own team.