Reserved Capacity vs On-Demand for a Predictable but Growing Workload
A SaaS company runs a steady-state API tier of 20 identical instances 24/7. Usage has grown 8% per quarter for six quarters and is expected to keep growing, but the growth is uneven across regions. The finance team is considering a three-year all-upfront reserved commitment for 20 instances to capture the deepest discount. Engineering warns that the workload may migrate to a newer instance family within 18 months as the runtime is upgraded. The company has a 12-month cash-flow constraint and cannot tie up large upfront payments. Current on-demand spend is roughly $30,000/month for this tier.
1. Evaluate the three-year all-upfront reservation against the company's constraints and recommend a commitment structure.
Read the worked answer
Three-year all-upfront conflicts with two stated constraints: the 12-month cash-flow limit (upfront payment) and the 18-month migration risk (a three-year term outlives the expected instance-family change). The deepest discount is real, but the risk-adjusted value is poor if the instances become obsolete before term end.
A better structure:
- One-year, no-upfront or partial-upfront reservations for the stable baseline (say 12–15 instances) to stay inside the cash-flow limit while capturing a meaningful discount.
- Convertible or family-flexible reservations if offered, so an instance-family migration does not strand the commitment. If only standard reservations exist, size the commitment to the portion of the fleet least likely to migrate.
- Leave the growth tail on on-demand or savings plans so the 8%/quarter growth is not over-committed; over-committing to a fixed count wastes money when growth outpaces the reservation.
Rule of thumb: commit only to the floor you are confident about for the full term, and keep the variable layer flexible. Re-evaluate at each 12-month mark.
2. Finance asks how to decide the reserved quantity given 8% quarterly growth. Walk through the reasoning for choosing a number rather than 'all 20.'
Read the worked answer
The reservation should cover the minimum fleet size you expect to run for the entire commitment term, not today's count. With 8% quarterly growth, today's 20 becomes larger over time, so growth is not the risk-migration and cash flow are. The risk is committing to instances you might not run for the full term.
Reasoning: estimate the floor. If the runtime upgrade could move, say, 30% of the fleet to a new family within 18 months, then only ~70% of today's 20 (about 14 instances) is a safe one-year commitment. Compare the per-instance discount against the break-even utilization: a one-year reservation typically pays off if the instance runs more than roughly two-thirds of the term. If you expect the floor to stay above that break-even, reserve it; otherwise stay on-demand.
Present it as: reserve the confident floor (≈14), leave the rest flexible, and revisit quarterly using actual utilization data rather than the growth projection alone.