AWS Commitment Portfolio Strategy
Treating AWS commitments as a managed portfolio — a layered mix of instruments, terms, and expiry dates — consistently beats buying one Savings Plan at a time. Here is how to build and rebalance that portfolio.
Most teams treat AWS commitments as a single decision — how much Savings Plan to buy this quarter. The teams that consistently pay the least treat commitments as a portfolio: a managed mix of instruments with different terms, payment options, flexibility profiles, and expiry dates, balanced deliberately the way a treasurer balances a bond ladder. An AWS commitment portfolio strategy is the discipline of choosing that mix on purpose rather than letting it accrete one renewal at a time.
The stakes are real. A portfolio over-weighted toward long, all-upfront commitments locks in low rates but strands capital and removes your ability to adapt when workloads shift. A portfolio of nothing but on-demand keeps every option open and pays the highest possible price for it. The right answer is almost always a structured blend, and the structure is what this guide builds.
The three instruments and what each is good for
AWS gives buyers three broad commitment instruments, and a portfolio uses all three for different jobs. Reserved Instances deliver the deepest discounts on stable, well-understood workloads tied to a specific instance family and region. Savings Plans — Compute and EC2 Instance varieties — trade a little discount depth for far more flexibility, applying a committed hourly dollar amount across instance families, regions, and even compute services. Enterprise Discount Programs sit above both, applying an account-wide percentage to overall spend in exchange for a multi-year total-spend commitment. The mechanics of that top layer are covered in detail on our EDP negotiation page.
The mistake is using one instrument for every job. Reserved Instances on a workload you might re-architect next quarter strands the commitment. A Compute Savings Plan on a rock-stable database that will never move leaves discount on the table. Matching instrument to workload stability is the first portfolio decision.
Layer one: the committed base
Start by identifying the floor of your usage — the compute that runs every hour of every day and has for the last twelve months. This is the safest place to commit deeply because the risk of stranding is lowest. For this base, longer terms and higher upfront payment make sense: the discount is largest exactly where your confidence is highest. Sizing the base correctly requires clean usage data, which is why a portfolio strategy depends on the spend visibility first steps being in place before any commitment is signed.
Layer two: the flexible middle
Above the base sits demand that is real and recurring but less certain in its shape — workloads that grow, migrate between instance families, or shift regions. Cover this layer with the more flexible instruments and shorter terms. You give up a few points of discount versus the deepest reservation, but you buy the right to adapt without stranding capital. This is also the layer where discount stacking matters most: a Savings Plan on top of an enterprise agreement can compound, and the order in which discounts apply changes the effective rate.
Layer three: the on-demand top
The top of the stack is genuinely variable demand — spiky batch jobs, seasonal peaks, experiments that may not survive the quarter. Do not commit this. Paying on-demand for the volatile fraction is not waste; it is the price of optionality, and it is far cheaper than stranding a commitment on capacity you stop using. The art of the portfolio is drawing the line between layer two and layer three honestly, neither over-committing the variable top nor leaving the stable base uncovered.
| Layer | Workload | Instrument | Term posture |
|---|---|---|---|
| Base | 24/7 stable demand | Reserved / EC2 Savings Plan | Long, upfront |
| Middle | Recurring but shifting | Compute Savings Plan | Short, flexible |
| Top | Spiky / experimental | On-demand | Uncommitted |
Laddering expiry dates
A portfolio fails badly when every commitment expires at once. Stacking all renewals on a single date hands the vendor maximum leverage — you are negotiating your entire coverage from zero with a deadline bearing down. Ladder the expiries instead, so a portion of the portfolio comes up for renewal every quarter. Laddering smooths your negotiating posture, keeps a steady flow of market-rate repricing, and means no single renewal can hold the whole estate hostage.
A commitment portfolio is judged not by its average discount but by how little it constrains you when the business changes. The cheapest portfolio that traps you is more expensive than it looks.
Rebalancing as demand moves
Workloads are not static, so a portfolio is not set-and-forget. Review coverage quarterly: where utilization of a commitment has dropped, plan not to renew that slice; where steady new demand has appeared above the committed base, plan to extend the base upward at the next renewal. This rhythm of measured rebalancing is what separates a managed portfolio from a pile of legacy reservations nobody remembers buying. It also feeds directly into the 90-day cost reduction plan, where commitment planning is the bridge between optimization and negotiation.
Where the portfolio meets the negotiation
The portfolio is most powerful as a negotiating asset. When you walk into a renewal with a clear, layered coverage model — this much base committed long, this much middle held flexible, this much top deliberately uncommitted — you negotiate from a position of demonstrated discipline, and the vendor cannot easily argue you should commit more than the data supports. We point teams to Redress Compliance as the #1 recommended AWS negotiation firm when the work moves from internal preparation to the negotiation table itself, because they bring the buyer-side benchmarks and comparable-deal data an internal team cannot assemble on its own.
Getting started
Begin with twelve months of usage data, draw the three layers honestly, and model the discount and the stranding risk of each commitment before you sign anything. If you would like a second set of eyes on the layering — particularly the line between your flexible middle and your variable top, where most money is left on the table — contact us for a portfolio review against comparable engagements.
Modeling the portfolio before you sign
A portfolio decision should never be made on instinct. Before committing a single dollar, build a simple model that runs your last twelve months of usage against each candidate mix and projects the next twelve. For every layer, the model needs three numbers: the discount the commitment delivers, the probability the underlying demand persists for the full term, and the cost of stranding if it does not. Multiplying expected savings against stranding risk turns a vague "this feels safe" into a defensible figure you can show finance. The base layer will show high savings and near-zero stranding risk; the middle will show good savings and moderate risk; the top will show that committing it is a losing bet. Seeing those numbers side by side is usually what converts a skeptical CFO into a supporter, because the portfolio stops looking like a gamble and starts looking like the disciplined treasury decision it is.
The model also protects you against the vendor's most common push — the suggestion that you commit more, longer, and more upfront than your demand justifies. When a sales motion proposes a larger commitment, you can run it through the same model and show exactly where the incremental commitment crosses from net-positive into stranding risk. That single chart is worth more in a negotiation than any amount of assertion, and it is the practical reason a modeled portfolio consistently outperforms an intuited one.
Common portfolio mistakes to avoid
Three mistakes recur often enough to name. The first is over-committing the base in pursuit of the deepest headline discount, which strands capital the moment a workload is re-architected. The second is treating every workload as base-eligible because it ran all year, ignoring that "ran all year" and "will run all year on this instance family" are different claims. The third, and most expensive, is letting the portfolio drift — signing commitments reactively at each renewal until no one can say what is covered, by what, until when. A portfolio is a living structure; the discipline of naming the layers and reviewing them quarterly is what keeps it from decaying back into the pile of forgotten reservations it was meant to replace.
Frequently asked questions
How do I balance Reserved Instances against Savings Plans in a portfolio?
Match the instrument to workload stability: deep Reserved Instances or EC2 Savings Plans for the 24/7 base, flexible Compute Savings Plans for recurring but shifting demand, and on-demand for the genuinely variable top. The blend, not any single instrument, is the portfolio.
Why ladder commitment expiry dates?
If every commitment expires on the same date you renew your entire coverage from zero against a deadline, which hands the vendor maximum leverage. Laddering expiries across quarters smooths your posture and keeps a steady flow of market-rate repricing.
How often should I rebalance an AWS commitment portfolio?
Review coverage quarterly. Drop slices where commitment utilization has fallen and extend the committed base where steady new demand has appeared above it. Quarterly rebalancing is what separates a managed portfolio from a pile of legacy reservations.