The cloud bill that convinced a company to buy its own servers

In late 2022, 37signals (makers of Basecamp and HEY) published a post announcing they were leaving the cloud. The number that anchored the argument: they'd spent roughly $3.2 million on cloud services in a single year — and their apps weren't even growing explosively. For a stable, profitable business, that felt less like a bargain and more like paying rent forever on something they could own.

The deeper insight isn't "cloud is expensive." It's that cloud pricing is really a premium for elasticity — the ability to spin servers up and down in seconds. That flexibility is genuinely valuable when your traffic is spiky or unpredictable (a startup that might 10x overnight, a Black Friday spike). But 37signals' load was steady and forecastable. They were paying the elasticity premium for flexibility they never used.

When demand is predictable, the math flips: buying hardware and amortizing it over years can be dramatically cheaper than renting equivalent capacity monthly. 37signals bought their own servers (roughly $600K in Dell machines) and projected saving around $7 million over five years. The trade-off is real — you take back the burden of capacity planning, hardware failures, and ops — but for the right workload, that burden costs far less than the premium.

This is why "cloud vs. own hardware" has no universal answer: it depends entirely on how uncertain your future demand is.