Startup Weighing CapEx for a Steady Web Tier
A 12-person startup runs a customer portal on three on-premises servers in a colocation cage. The cage costs $1,400/month including power and a 100 Mbps commit. The servers are four years old and need $9,000 of refresh in the next quarter. Traffic is flat at roughly 40 requests/second with a predictable weekday peak of 90. The CTO wants to move to AWS but the CFO argues that ownership is cheaper because the hardware is already paid for. Finance requires a three-year view and a 10% hurdle rate. The team has no dedicated operations staff and the lead engineer spends about six hours a week on patching, backup verification, and disk replacement. Management has asked for a recommendation that separates fixed from variable cost and states what would have to be true for each option to win.
1. Explain how the team should frame the CapEx-versus-variable-cost comparison, and what non-cost factors should be weighed.
Read the worked answer
The comparison must put both options on a three-year, fully loaded basis. On-premises cost is mostly fixed: cage fees ($1,400 x 36 = $50,400), refresh ($9,000), and the engineer's six hours a week. At a loaded rate of, say, $75/hour that is about $23,400 over three years, plus the risk of a second refresh inside the window. AWS cost is variable and elastic: a small always-on footprint for the baseline 40 rps plus burst capacity for the 90 rps peak. The key judgement is that flat, predictable demand does not automatically favor ownership, because the startup is paying for peak capacity 24/7 while its utilization is low. Non-cost factors matter here: no dedicated ops staff means patching and backup verification are a real risk, and a failed disk during a weekday peak has revenue impact. Conversely, AWS introduces new skills and the risk of uncontrolled spend without budgets and alerts. The recommendation should state assumptions explicitly: if demand stays flat and the team is willing to keep operating hardware, ownership can win on pure cash; if demand grows, spikes, or the engineer's time is better spent on product, variable cost wins.
2. The CFO asks what would have to be true for the cloud option to be cheaper. Give the reasoning, not just a number.
Read the worked answer
Cloud wins on cost when the avoided fixed cost plus the value of elasticity and freed engineering time exceeds the variable bill. Concretely, the on-premises baseline is roughly $50,400 in cage fees plus $9,000 refresh plus $23,400 of labor, about $82,800 over three years, or roughly $2,300/month. For AWS to beat that, the steady-state footprint must come in under that figure after accounting for the fact that you no longer pay the cage or the refresh. That is plausible for a small web tier, but the answer must be conditional: it holds only if (1) demand does not require large always-on instances, (2) the team uses budgets, cost anomaly detection, and right-sizing so the variable bill does not drift, and (3) the six hours a week is genuinely redeployed rather than replaced by a different six hours of cloud administration. If the team instead over-provisions for the 90 rps peak around the clock, the elasticity advantage disappears and the comparison narrows. The honest answer is a break-even model with sensitivity to utilization and growth, not a single verdict.