Your Real AWS Discount After Savings Plans: Why the Headline Percent Overstates It

AWS applies your private pricing discount to what is left after Savings Plans and Reserved Instances already cut the rate, so a 12 percent PPA on a 66 percent Savings Plan is worth roughly 4 points, not 12. This page gives you the arithmetic, the benchmark denominator, and the concession list to demand once you can prove what the discount is actually worth.

The Discount You Signed Is Not the Discount You Got

Somewhere in your business case there is a line that reads like this: 66 percent from a three-year Compute Savings Plan, plus 12 percent from the private pricing agreement, equals 78 percent off on-demand. That number is wrong by roughly 8 points of on-demand equivalent, and the error is not a rounding issue. AWS applies the private pricing discount to what remains after the Savings Plan has already cut the rate, so the real landing zone is 70.1 percent. On a $10M annual on-demand-equivalent estate, the gap between the modeled 78 percent and the delivered 70.1 percent is about $790K a year, or $2.4M over the term you just signed. Finance approved the deal on the first number. Procurement will be measured against it.

Treat this as a negotiation failure rather than an accounting curiosity, because of what you traded for the overstated value. To get that 12 percent you almost certainly accepted a three-year term instead of one, a commitment that ratchets upward every year and never down, an opening commit that in most engagements we have advised sat 15 to 30 percent above the customer's realistic spend curve, and in many cases mandatory Enterprise Support priced as a percentage of gross spend. You paid full price in concessions for something worth about a third of what it looked like. AWS knew the arithmetic. Your model did not.

You paid full price in concessions for a discount worth about a third of what it looked like on the slide.

The position for everything that follows is simple. Stop benchmarking the PPA percentage in isolation, because in isolation it is a marketing number. Benchmark the effective rate off on-demand, price every concession against the incremental points the PPA actually adds, and go back to the table with the delta. Before you accept AWS's tier framing, check your commit level against published AWS private pricing discount bands by spend tier, then decide whether the term length is buying you anything real.

The Multiplicative Stack: Effective Rate = 1 − (1 − SP%) × (1 − PPA%)

There is a definitional argument in the market that wastes buyers' time. Some advisors tell you EDP and PPA discounts do not stack with Savings Plans at all. AWS's own re:Post guidance says the two are applied sequentially rather than additively. Both descriptions produce identical bills. Sequential application on a shrinking base is mathematically the same thing as saying the second discount only bites what the first one left behind. The formula settles it: Effective rate = 1 − (1 − SP%) × (1 − PPA%). Put that line in the business case and the dispute disappears.

The precedence order inside the bill matters when you build the model. Reserved Instances apply first where you still hold them, then EC2 Instance Savings Plans, then Compute Savings Plans, then the private pricing layer lands on whatever residual is left across the bill. Run the flagship case. A three-year all-upfront Compute Savings Plan at the published 66 percent ceiling leaves a residual of 34 cents on the dollar. A 12 percent PPA applied to that 34 cents removes 4.1 cents. Your effective rate is 70.1 percent, and the PPA contributed 4.1 points against a headline of 12. That is roughly one third of face value. The deeper your Savings Plan coverage, the less your negotiated percentage is worth, which is precisely the inversion AWS never volunteers in the deal review.

Find your cell in the table below. The Savings Plan axis spans the realistic band of 32 to 72 percent rather than the marketing ceilings, because no-upfront and one-year terms sit well below 66 percent. Each cell shows the effective rate off on-demand, with the incremental points the PPA actually delivers in parentheses.

Read the right-hand columns carefully. At 72 percent EC2 Instance Savings Plan coverage, moving your PPA from 10 percent to 20 percent, a doubling that AWS will price as two full commitment tiers and years of additional term, buys you 2.8 incremental points. On a $10M on-demand-equivalent estate that is $280K a year. Ask what AWS wants in return: a commit ramp from $5M to $13M, a five-year lock, exit exposure that can exceed 75 percent of remaining unused commit. The arithmetic tells you whether the trade clears. Before you concede term length, quantify what the extra years are worth using the one-year versus three-year discount premium benchmark, and compare it against the same 34-cent residual. In most cases, service-level adders on specific high-spend services return more incremental points per concession than another two points on the cross-service rate.

Your Denominator Is Wrong Too: Realised Versus Contracted Savings Plan Rates

Once you accept that the private pricing layer only bites on the residual, the next problem is that the residual itself is usually smaller than your model assumes. AWS quotes ceilings: Compute Savings Plans at 66 percent, EC2 Instance Savings Plans at 72 percent, SageMaker at 64 percent, and the Database Savings Plan (launched December 2024) at a much thinner 35 percent. Every one of those numbers assumes three years, all upfront, and near-perfect coverage. Shift to one year or no upfront and the realised rate drops materially; the practical band across the portfolio runs closer to 32 to 72 percent. That matters because the effective-rate arithmetic is only as honest as the Savings Plan input. If you feed it 66 percent when your realised blended rate is 48 percent, you have built a business case on a number your billing data will not support at renewal, and AWS will happily let you present it.

The forfeiture rule is what separates contracted from realised. Any committed dollar per hour your usage does not cover that hour is gone, permanently, with no carry-forward. So a plan sized for peak capacity and running at 82 percent utilisation is not delivering 66 percent; it is delivering roughly 54. This is also where the Compute versus EC2 Instance choice stops being a technical preference and becomes a negotiation input. Six extra points on a $500K annual EC2 bill is worth $30K, real money, but that entire advantage disappears if the tighter instance-family lock drops utilisation to 70 percent. The trade between Savings Plans and Reserved Instances should be priced on realised savings across the last four quarters of billing data, not on the marketing ceiling.

The GPU distortion breaks the model outright for AI-heavy estates. In advisory work, GPU instances attract roughly 15 to 25 percent less discount than comparable CPU capacity at the same commitment tier. If half your incremental spend is accelerated compute, benchmarking your total agreement against a blended 66 percent ceiling means you are measuring yourself against a rate the portfolio structurally cannot reach. Build the denominator by workload segment (CPU steady-state, GPU training, GPU inference, managed services, egress), compute realised savings for each from your Cost and Usage Report, then blend. That number, not the AWS slide, is the base your private pricing percentage sits on.

The Offsets AWS Never Puts in the Model: Support, Ratchets, and Shortfall

Even a clean 70.1 percent effective rate is a gross figure. Enterprise Support is calculated on your pre-discount usage in most agreements and lands as a direct offset: 10 percent of the first $150K of monthly spend, 7 percent to $500K, 5 percent to $1M, 3 percent above that, with a $15K monthly floor. On $3M monthly spend that is roughly $93,500 a month, about $1.12M a year, and it arrives as a line item the account team never puts in the savings deck. Then there is the retirement gap. Discount value does not consume commitment. Spend $2M gross, receive $200K in private pricing benefit, and only $1.8M retires against your commit. Over a five-year ramp that gap alone can push a well-behaved customer into shortfall.

Price the downside before you sign, not after. The true-up is not a penalty in AWS's language, it is invoiced commitment, and it is payable whether or not you used a dollar of capacity. The one-way ratchet compounds it: next year's commit cannot fall below this year's, so a single over-committed year sets a permanent floor under every subsequent renewal. Advisory experience shows AWS's opening commitment typically sits 15 to 30 percent above the customer's realistic spend curve, which is precisely the mechanism that converts a discount into a shortfall charge. Insist on shortfall carry-forward into the following year, a downward flex band of at least 10 percent on the ratchet, and termination for convenience capped at unused commit for the current year only.

A 70 percent headline rate that pays $1.12M in support, retires only 90 cents of commitment per dollar spent, and carries an $800K true-up exposure is not a 70 percent deal.

The number your CFO should approve is net of all of it: effective rate after the multiplicative stack, less support as a percentage of gross spend, less expected shortfall probability-weighted across the ramp. On the example above that lands nearer 58 to 62 percent. Take that figure into the room alongside the published discount bands by spend tier and ask AWS to explain the gap.

Where the Leverage Actually Sits Once You Fix the Math

Once the model shows that a 12 percent private pricing rate is worth roughly 4 points against on demand on committed workloads, the negotiation target has to move. Buyers spend most of their concession budget buying PPA points, and the currency they pay in is term length and commit growth, exactly the two things AWS wants most. That trade is now visibly bad: a fifth year and a 20 percent annual ramp on a $5M base commit exposes tens of millions of dollars of future obligation to buy an increment worth about a third of its face value on the covered portion of the bill. Reprice the ask. The residual, the portion of spend that Savings Plans and RIs never touch, is where the PPA layer still lands at full strength, and that is where the concessions should be concentrated. Push for service-level discount adders on the services outside Savings Plan coverage, S3, data transfer, managed databases and analytics, where an incremental point is a real point. Force Marketplace spend to retire commitment at 100 percent, not a partial rate, because that is worth more than a PPA point on most portfolios. Attack Enterprise Support: the $10K per month floor for customers above $5M annual spend has been conceded in our engagements, though rarely, and it survives outside the discount stack entirely. Then cap the uplift. Flat or capped annual growth is worth more than a point of discount because it removes the shortfall exposure that turns a discount into a true-up. Use the discount bands by spend tier to establish what your commitment level should return, and the one-year versus three-year premium to price the term you are being asked to sell. Our forthcoming work on commit tier breaks, on above-band discounts and the concessions AWS charges for them, and on uplift cap benchmarks sharpens each of these into a target number.

What AWS Will Say When You Present the Effective Rate

Expect four counters, in roughly this order. First, that your benchmark is not comparable because every customer's service mix, region footprint and commit profile differs. That is true and irrelevant: the on demand equivalent rate is mix-neutral by construction because it divides what you actually paid by what the same usage would have cost at list. Hold that as the single scoreboard metric for the entire negotiation and refuse to score anything else. Second, the pivot from rate to total dollars saved. AWS will show you a large absolute savings figure, most of which your own Savings Plan coverage generated. Answer it by splitting the number: this much came from commitment instruments we bought and manage, this much came from your private pricing layer, and the second figure is the only one under negotiation. Third, the offer of one more PPA point in exchange for a fifth year or a 20 percent annual ramp. Price it in dollars before you respond. On a $5M Year 1 commit ramping at 20 percent, the fifth year alone carries roughly $10M of obligation, and one incremental PPA point on covered workloads is worth about a third of a point against on demand. Say the number out loud in the room. Fourth, the reframe that Savings Plans are your own optimization work and therefore not AWS's discount to be netted against. Concede the premise and reject the conclusion: whoever did the work, the list price rate you are paying is the only number that determines the invoice, so it is the only number that determines whether the deal is competitive. On commit sizing, hold hard. In advisory engagements the opening commitment routinely sits 15 to 30 percent above the customer's realistic spend curve, which manufactures a shortfall true-up at term end and quietly converts your discount into a penalty. Model your own curve, present it, and refuse any Year 1 number above it. The strong outcome: capped uplift, Marketplace retiring at full value, adders on the uncovered residual, and a commit you can actually consume.

Do this first: rebuild last quarter's bill as an on demand equivalent rate, then bring that single number to the next call and make AWS negotiate against it.

Build the Model and Reset the Ask This Week

Start with data, not a counteroffer. Pull twelve months of line-item spend from Cost and Usage Report, split three ways: Savings Plan covered, RI covered, and on-demand. Compute realised SP savings, meaning actual dollars saved divided by what the same usage would have cost at on-demand, not the 66 or 72 percent ceiling on the purchase order. In our engagements the realised figure typically lands several points below the contracted rate because unused committed dollars per hour are forfeited hourly. Then apply the formula: effective rate equals 1 minus (1 minus realised SP percent) times (1 minus PPA percent). Subtract Enterprise Support, which is billed on pre-discount usage, and subtract your probability-weighted shortfall exposure. That net number is your real discount, and it is the only number worth putting in a business case.

Now restate the ask in those terms. Do not ask for "more discount." Ask for a target effective rate on on-demand equivalent spend, a capped annual uplift (single digits, not the 15 to 30 percent above realistic curve AWS typically opens with), a Year 1 commit set at or below your modelled low case, and Enterprise Support discounted or credited separately. Benchmark the PPA layer against published discount bands by spend tier so you know whether 12 percent is generous or lazy at your volume.

Expect AWS to defend the headline. Your fallback if the residual will not move: shorten the term. Trade the five-year ratchet for a two-year deal and revisit, using the term premium arithmetic to price what you are giving up. A strong outcome is 3 to 5 points of genuine effective-rate improvement plus a commit you will actually consume. Build the model before the next call.

Related advisory: the full topic guide.

Frequently asked questions

Do AWS EDP discounts stack on top of Savings Plans?

They apply sequentially, not additively. Reserved Instances and Savings Plans reduce the rate first, then the private pricing discount is applied to what remains. A 12 percent PPA on top of a 66 percent Savings Plan produces roughly 70 percent off on-demand, not 78 percent.

What formula should I use to calculate my real AWS discount?

Effective rate = 1 − (1 − SP%) × (1 − PPA%). Use realised Savings Plan savings rather than the published ceiling, because term length, payment option, and forfeited committed hours all reduce the actual rate you achieve.

How much is a PPA discount point actually worth on heavily committed spend?

On workloads already covered at 66 percent, each PPA point delivers about 0.34 points of on-demand equivalent savings. On uncommitted on-demand spend it delivers the full point, which is why the mix between committed and on-demand spend determines how hard you should fight for PPA percentage.

Does Enterprise Support reduce my effective AWS discount?

Yes, and materially. Enterprise Support is mandatory under an EDP and runs from 10 percent of the first $150K of monthly charges down to 3 percent above $1M, with a $15K monthly minimum. A customer at $3M per month pays roughly $1.12M annually, which should be netted against the discount before the business case is approved.

Does my PPA discount count toward my annual commitment?

No. The discount value does not retire commitment. If you spend $2M and receive $200K in discount benefit, only $1.8M counts against your commit, which is a common cause of unexpected shortfall true-ups at year end.

What discount range do buyers actually achieve on AWS private pricing?

Observed outcomes typically run from about 5 percent to about 20 percent on net spend, with $1M to $3M commits landing at 8 to 12 percent, $3M to $10M at 10 to 16 percent, and $10M to $25M reaching 14 to 20 percent. The top of market requires very large commitments and long terms, and AWS does not publish the tiers.