Term length
56%
The three-year term carries the deepest discount.
Committing for three years instead of one buys a meaningfully deeper discount, and that holds across every service in this catalog: RDS, ElastiCache, and EC2 alike.
A fleet that's going to keep running for years is giving up real savings by defaulting to the shorter term out of caution.
The cautious term
36%
Even the one-year term still discounts well below on-demand.
A shorter commitment is the safer bet for a fleet that might resize or retire soon, and it still locks in a real discount against paying on-demand rates the whole time.
Every reservation term is worth pricing out. The only question is how much certainty you're willing to trade for the deeper of the two discounts.
A gap in the matrix
0 of 1840
Amazon RDS simply doesn't sell a three-year Reserved Instance with nothing paid upfront.
Every RDS record in this catalog was checked against AWS's own price list, across every region and engine tracked here, and that combination never turns up. ElastiCache and EC2 both offer it.
Budgeting for a cash-free, three-year RDS commitment isn't a planning gap to close. AWS never offered that purchase option in the first place.
Payment option
87%–91%
Paying nothing upfront still keeps most of the three-year discount.
Where AWS does sell a three-year Reserved Instance with no money down, its discount sits close behind the fully prepaid version of the same offer. Cash up front buys a further edge, not a doubling of the saving.
A team that can't stomach a large upfront payment isn't giving up the bulk of the discount by choosing not to pay one.
Which lever matters more
20 vs 6 pts
Choosing a longer term moves the discount far more than choosing how you pay for it.
Stretching a commitment from one year to three shifts the discount by a wide margin. Switching from no money down to paying it all in full moves the number by comparison, but nowhere near as far.
Anyone agonizing over the payment-option dropdown is tuning the smaller lever. Term length is where the decision that actually matters lives.
Region matters
up to 16 pts
The same instance type can carry a very different discount depending on the region it's reserved in.
Across the regions this catalog tracks, most instance types see only a narrow gap between their best and worst regional discount, but the outliers move far more than that, on the exact same class, engine, and deployment option.
A discount quoted for one region isn't a safe stand-in for another; the region a fleet actually runs in belongs in the math, not left as an afterthought.