AWS caps private pricing discounts at 5 to 20 percent, so the 4 to 7 points you gain over their opening offer come from eight clauses in the addendum, not from the discount table

Across roughly 45 to 60 AWS renewals benchmarked from 2024 to 2026, the median final discount landed 4 to 7 points above the account team's opening proposal, and almost none of that movement came from arguing about the percentage. It came from commit-eligibility definitions, service exclusions, price protection scope, Marketplace retirement caps, and off-ramps. If you spend your negotiation cycle on the headline band, you will sign a number that looks fine and a contract that costs you 15 to 25 percent of total value over three years.

Prepared by Redress Compliance · August 22, 2026 · AWS private pricing advisory. EDP and PPA negotiations benchmarked 2024 to 2026.

Start by dropping the vocabulary argument. AWS deprecated the Enterprise Discount Program label and now presents essentially every commercial offer as a Private Pricing Addendum, but the commitment structure, the tier logic, and the Marketplace eligibility rules are unchanged.

If your last contract said EDP and the renewal paper says PPA, only the letterhead moved. That matters because your prior-cycle redlines are still live ammunition, and because AWS reps occasionally use the rename to suggest the terms are "new standard paper" and therefore less negotiable.

They are not.

What you are actually redlining is a short addendum sitting on top of the AWS Customer Agreement, and inside it sit eight clause families that carry every dollar of real value: the definition of Commitment-Eligible Fees, price protection scope, service exclusions and substitution rights.

The Marketplace retirement cap, support fee treatment, assignment and change of control, termination and off-ramps, and confidentiality plus benchmarking restrictions.

The discount percentage is one line. These eight are the contract.

The published AWS addendum language is worth reading before you table anything, because it tells you exactly where the definitional traps live.

Commitment-Eligible Fees in a real signed AWS PPA (the UK DVLA agreement, discount term April 2024 to March 2027) reads as fees incurred under Eligible Accounts, excluding fees paid by applying Available Balance, excluding taxes, and net of applicable discounts.

Read that carefully: credits AWS gives you do not retire your commit, and the discount AWS gives you shrinks the fees that count toward the number you promised to hit. Both effects run against you, and both are drafting choices, not laws of physics.

Approach this the way a disciplined buyer approaches any clause-driven enterprise deal, closer to the method in the discount is on the order form, the money is in the clauses framing than to a discount-table conversation.

Accept the arithmetic reality up front: AWS private pricing discounts run 5 to 20 percent, stepped by commitment threshold, and they do not go above 20 percent no matter how the ask is framed. A $1M to $3M annual commit lands at roughly 8 to 12 percent.

A $5M to $10M commit lands at roughly 12 to 16 percent. When your account team opens, they are already inside a point or two of the ceiling for your spend level, because the band is set by tier policy, not by the rep's goodwill.

This is why the buyer who spends the cycle arguing for 22 percent loses. There is no 22 percent. You are negotiating against a wall, and while you push on it, the clauses that actually move money go unredlined.

The 4 to 7 points of median improvement over the opening proposal come from three places, none of which appear in the discount table.

First, service-specific pricing stacked on the heaviest compute and managed-services lines, which in upper-enterprise estates has added meaningful incremental value on top of the tier discount rather than replacing it.

Second, widening what retires the commit: every dollar of Marketplace spend, support fee, or credit-funded consumption you pull inside Commitment-Eligible Fees is a dollar you no longer have to duplicate in raw AWS consumption to hit your number. Third, shifting offer composition.

AWS builds offers as a mix of credits and discounts, and the mix changes quarter to quarter based on what the field is funded to give. Credits expire, do not retire commit, and are worth materially less than the same face value in discount.

Converting even a third of a credit-heavy offer into structural discount is worth more than two points on the headline.

Know what AWS does when you press on any of this.

The reliable response is to reframe your ask as a term problem: "we can get you closer to that number on a five-year." That is the trade being offered.

And it is usually a bad one, because you are paying for two or three points with your entire renewal-cycle leverage and your ability to reprice against a market where AWS margin is expanding and capacity is scarce.

Hold at one to three years. Tell them the term is fixed and the value has to come from eligibility scope and service-level pricing instead. That single refusal is what forces the conversation onto the ground where you actually win.

The published AWS Private Pricing Addendum language is short and it is doing a lot of work: Commitment-Eligible Fees are fees incurred under Eligible Accounts, excluding fees paid by applying the Available Balance, excluding taxes, and net of any applicable discounts.

Read that as three separate reductions to the denominator you get measured against. Eligible Accounts decides whose spend counts. The Available Balance carve-out means credits AWS hands you, including the incentive credits inside the offer composition, burn spend without retiring commit.

And "net of applicable discounts" is the one that catches finance teams cold: a larger discount mechanically slows retirement. On a $10M annual commit, a 15 percent discount means you consume roughly $11.8M of list-priced usage to retire $10M of commit; at 20 percent you need about $12.5M.

The account team will present the higher discount as pure win. It is a win on unit price and a tightening of the shortfall test at the same time.

The Eligible Accounts scope is where most real money moves. AWS opens with a schedule of named account IDs, frozen at signature.

Every acquisition, every newly onboarded subsidiary, every dev estate that a business unit spun up outside central IT then sits outside the commit unless someone remembers to file an amendment.

In practice that is how a company running $14M of actual AWS spend lands in a shortfall conversation on a $12M commit. Push for named accounts plus all majority-owned affiliates and any entity acquired during the term, with automatic inclusion on notice rather than AWS consent.

If AWS resists automatic inclusion, the fallback is a written 30-day add process with no re-pricing trigger, so adding accounts cannot reopen your rate card. The same pattern shows up in Snowflake commit measurement, and buyers who fought it there should reuse the language.

The third ask is procedural and AWS rarely fights it hard: a quarterly reconciliation statement showing Commitment-Eligible Fees against commit, with a 60-day dispute window and a written correction path.

Without it you discover a measurement disagreement in month 34, when you have no leverage and no time. Our commit measurement subpage works the mechanics line by line.

Do the arithmetic before the first pricing call: take your trailing twelve months, strip taxes, strip anything paid from credits, apply the proposed discount, and that is your true retirement base. Bring that number to the table as the ceiling on commit, not the floor.

Standard PPA price protection reads generously and delivers narrowly. What you typically get is protection on on-demand public rate cards for in-scope services: AWS will not raise the list price of the services in your schedule during the discount term, or if it does, your effective rate is held.

What you almost never get without asking is protection on reservation and capacity products, meaning Reserved Instances, Savings Plans, capacity blocks, and the reserved GPU and accelerator constructs.

That is the exact category where 2026 to 2027 cost inflation is landing, and the exclusion is not an accident of drafting.

The evidence for intent is on the record. Amazon raised 2026 capex guidance to roughly $220B from about $200B, explicitly citing higher memory chip and component costs, and Jassy said AWS will not have enough capacity to meet demand in 2026 and expects the same in 2027.

Asked directly about pricing under input-cost inflation, management confirmed that new deals are priced with current costs incorporated.

Translate that into contract terms: AWS is telling you it will pass component inflation through, and the only surface it can pass it through on, if your on-demand rates are locked, is the reservation and capacity products your rate protection does not cover.

A 37 percent growth quarter with a 39 percent segment operating margin means AWS is not short of room to absorb a protection clause. It is short of capacity, which is a different problem and a different negotiation.

The concrete ask has two forms and you should table both.

First, extend rate protection to the named reservation and capacity products you actually buy, listed by SKU family in the addendum, not by generic reference to "AWS Services." Second, where AWS refuses extension, insist on a stated cap on year-over-year increases for the excluded categories.

In our experience across recent enterprise negotiations, a cap in the 3 to 5 percent range per contract year is defensible and gets agreed more often than a flat freeze, because it lets AWS argue it preserved pricing flexibility.

Add a third fallback that costs AWS almost nothing to give and protects you materially: if a covered service is renamed, re-SKU'd, or superseded by a successor offering, protection follows the successor. Without that line, a service refresh legally exits your protection scope.

Model the exposure so the ask has a number behind it. If reserved capacity is 40 percent of a $12M annual spend and unprotected rates move 8 percent in year two and again in year three, you are absorbing roughly $800K of unbudgeted cost across the term.

That figure, not a principled argument about fairness, is what moves an account team that is measured on committed revenue rather than realized rate. Our price protection gap subpage runs the category-by-category exposure math.

A 36-month commit priced against your 2026 service mix is, in substance, a bet that your architecture stays still for three years. It will not.

The estates we benchmark re-platform something material roughly every 12 to 18 months: EC2 to Graviton, self-managed Kafka to MSK, EMR to Athena or serverless Spark, Redshift to a lakehouse pattern, and increasingly a wholesale shift of budget toward Bedrock and inference-heavy workloads.

The commit itself usually survives that, because most retirement definitions are broad. What breaks is the service-specific discount schedule in the addendum, where the 11 to 19 points of stacked discount that made your business case sit on named SKUs.

Migrate off the named SKU and those points evaporate while the commit stays whole, which is the worst combination available: you keep the obligation and lose the price.

AWS will not volunteer that asymmetry, and the account team genuinely may not model it, because their comp is tied to consumption, not to your blended rate.

Table three things. First, substitution rights: if a workload moves from a discounted service to a functionally equivalent AWS service, the discount follows the workload at no less than the original rate.

Second, treat every service-specific rate as a floor, not a schedule: if AWS publishes a lower public price or a better tier during the term, you take the better of the two, which stops the discount from being quietly overtaken by list price movement.

Third, cover deprecation and successor SKUs explicitly: when a discounted service is retired or superseded, the successor inherits the same discount automatically rather than reverting to list.

Our subpage on swap and substitution rights walks the specific drafting; the same architectural logic applies when you are negotiating reserved capacity contracts on the other hyperscaler, where committed units are even more tightly bound to named SKUs.

Expect pushback framed as operational complexity: AWS will say substitution cannot be administered and will counter with an annual review instead. An annual review is a conversation, not a right.

Hold for a named list of five to eight services with swap rights, 30 days written notice, and a no net reduction in blended discount guarantee.

A strong outcome protects 2 to 4 points of blended rate across the term, which on a $10M annual commit is $600,000 to $1.2M you would otherwise hand back through re-platforming you were going to do anyway.

The discount band is genuinely capped at 5 to 20 percent, and AWS defends that ceiling because it is a global consistency story told to every other customer.

What is not capped, and what the account team has real internal latitude on, is the composition of the offer: how much arrives as a discount percentage against run-rate spend and how much arrives as credits, migration funds, POC funding, or service-specific promotional balances.

That composition changes quarter to quarter, and it is the single largest source of variance between two customers who both signed "17 percent."

Credits and discount points are not interchangeable, and the gap runs in AWS's favor on four axes. Credits expire, usually inside 12 to 24 months, and unused balances are simply revenue AWS never had to give up.

Credits do not compound into your run-rate base: a discount point applies to every dollar you spend for the whole term and every dollar you spend after it, while a credit applies once.

Credits are frequently service-restricted, so the funding is available for the workload AWS wants you to start rather than the workload you actually run. And most consequentially, credits do not move your renewal baseline. When you come back in 36 months, AWS prices from your net effective rate.

A customer who took 12 points plus $4M in credits renews from a 12-point base. A customer who took 16 points renews from 16. That difference compounds across two renewal cycles and is worth multiples of the credit's face value.

AWS's incentives point at credits for structural reasons, not tactical ones.

Credits are booked differently, they are frequently funded from program pools rather than the deal's own margin, and they let a rep close inside a quarter without seeking approval for a discount point that would breach a tier.

That is why credit-heavy offers cluster at quarter end and fiscal year end. It is also why a credit-heavy offer often looks generous: the face value is large precisely because AWS is confident a meaningful fraction will never be consumed.

A buyer who models credits at face value overpays systematically. The fix is a single blended-rate comparison that forces every offer variant onto one axis.

Take your three-year forecast spend, apply the discount to the full base, then haircut credits by expected consumption (we use 55 to 75 percent based on prior-cycle burn, not on AWS's projection), discount the remainder for time value.

And exclude any credit tied to a service you have no funded plan to deploy.

Divide total value by total spend. That is your real number. Run it for AWS's opening offer, for their revised offer, and for your own counter, and the ranking usually inverts at least once.

The same discipline applies wherever discount and non-cash value are mixed together, which is why the pattern shows up in Snowflake order forms where the headline sits above the real economics as well.

The counter that works is not "give us more discount." It is "we will accept your total value envelope, convert it to rate." Handing AWS an unchanged deal value while changing its form is a small ask internally and a large gain for you.

Where a rep cannot move on percentage, they can often move 30 to 50 percent of the credit pool into rate, which on a $30M three-year deal is typically 1.5 to 3 blended points.

Failing that, extend credit expiry to the full term, strip service restrictions, and secure written confirmation that renewal pricing anchors to the discount rate, not the net effective rate including credits.

These two items get conflated in almost every deal room I have sat in, and AWS benefits from the confusion. The Marketplace contribution cap is not a discount mechanism at all.

It is a retirement lever: roughly 25 percent of your annual commitment can be satisfied by third party software purchased through AWS Marketplace, and that percentage is a term in your addendum, not a law of physics.

I have seen it written at 30 and at 35 percent for accounts with heavy ISV footprints. The second carve-out runs the other way.

Enterprise Support fees are typically calculated on undiscounted usage and sit outside the discount entirely, which on a $10M annual commit means a support line in the range of $300K to $700K per year that your negotiated percentage never touches.

Neither of these is on the discount table your account manager brings to the first meeting, and neither shows up in the total-cost model finance signed off on.

The May 2025 narrowing is the part most renewal teams miss. AWS tightened Marketplace eligibility to SaaS offerings fully deployed on AWS infrastructure, which knocked a slice of previously eligible ISV spend out of retirement scope overnight.

If your prior term relied on Marketplace to close a $2M gap in year three, re-validate every vendor before signature, not after. Ask for the eligible-vendor list in writing, name by name, with the AWS Marketplace listing ID attached.

AWS will resist producing a static list because eligibility is administered dynamically and they do not want to be held to a snapshot. That resistance is the point: the risk of a mid-term reclassification is currently yours, and moving it is worth more than a point of discount on a mid-size commit.

The realistic ask on the cap is 30 percent as a floor with a written mechanism to revisit at 35 percent if Marketplace volume exceeds a stated threshold. AWS concedes this most readily when your Marketplace pipeline is real and named, because Marketplace revenue carries strategic weight internally.

On support, table two questions: does Enterprise Support count toward commit retirement, and is it subject to the discount. The honest answer from AWS on the second is usually no.

Get the first in writing anyway, because retirement credit on a $500K annual support line is worth more than arguing about a discount you will not win.

The same clause discipline shows up in adjacent cloud commitments, and the capacity reservation logic in Azure OpenAI PTU deals is a useful comparison for how buyers price a commitment they cannot fully control.

These clauses are worthless until the day they are the only thing that matters.

AWS's default addendum position is straightforward and entirely in their favor: the negotiated discount does not automatically survive a change of control, there is no termination for convenience, and if you fall short of your commitment you pay the difference.

Signed as drafted, that means an acquisition, a divestiture, or a strategic pivot away from a workload converts your pricing agreement into a liability.

I have watched a buyer discover, six weeks into diligence, that their EDP discount was not assignable and the acquirer's own AWS agreement would not absorb it. The shortfall exposure was disclosed as a contingent liability and it moved the purchase price.

Know what AWS will actually concede. Assignment to an acquiring entity with prior written notice is normal and usually granted, sometimes with a consent-not-unreasonably-withheld formulation.

Divestiture carve-outs are also achievable: language that allows a divested business unit to take a proportional slice of the commitment with it, or that reduces your remaining commit by the divested entity's trailing twelve months of eligible spend.

That second construction is the one to fight for, because it converts an open-ended risk into an arithmetic outcome. AWS will not give you unilateral termination for convenience without a shortfall payment. Stop asking for it after the second round and spend the leverage elsewhere.

What they will do is define the shortfall formula precisely, and that is where the money is.

Price the exposure explicitly rather than leaving it open. A commitment shortfall on a $10M annual commit, in a bad year, is a seven figure invoice with no service delivered against it.

Insist the addendum states the shortfall calculation, whether unused commitment rolls forward, and whether the shortfall is payable at list or at your discounted rate (the difference matters and AWS will default to list if you do not specify).

Then model it: what does an 80 percent attainment year cost, and does the discount you gained cover it.

The assignment and termination provisions warrant their own line-by-line treatment during redline, and the detailed subpages on both clauses work through the specific language to table. Treat the off-ramp as a priced option, not a legal formality.

The confidentiality clause in a standard AWS PPA is drafted to make your next negotiation harder, and it costs AWS nothing to defend, which is exactly why it is the cheapest clause on the page to fix.

As written, most versions bar disclosure of pricing, discount rates, and commitment structure to any third party without prior written consent.

Read literally, that captures the advisory firm you retain to benchmark the deal, the peer CIO you would call to sanity check a 14 percent band, and the anonymized data pool that would tell you whether AWS gave a comparable estate 4 points more.

AWS is not being sinister here, it is being rational: information asymmetry is the discount band's structural support. If nobody in your peer group can compare notes, the 5 to 20 percent range stays a fog rather than a distribution you can point at.

Table three carve-outs and expect two of them to land without escalation, in our experience across recent renewals: disclosure to legal, tax, and audit advisors, disclosure to retained third-party negotiation advisors under equivalent NDA.

And participation in aggregated anonymized benchmarking where your name is not attached to the numbers.

AWS account teams rarely burn political capital defending this clause because it is not a revenue term to them, and the deal desk has no line item for it.

The compounding effect is what makes it worth the redline: sign it three cycles running and you enter your 2032 renewal with no comparative evidence, arguing percentages against a vendor holding every data point.

Our detailed treatment of the confidentiality and benchmark clause sets out the exact insert language. It is the same discipline we apply to discount versus clause value in consumption contracts.

AWS will offer you discount points for a five-year term, and the trade is worse than it looks. Jassy said on the record that AWS will not have enough capacity to meet demand in 2026 and expects the same through 2027, with capex raised to roughly $220 billion.

A vendor that cannot serve its pipeline is not desperate for your 2031 spend, it is buying certainty cheaply. Meanwhile your renewal is the only recurring event where you hold anything.

Five years means one leverage event instead of two, across a period in which memory inflation is already being priced into new deals and your architecture will not resemble today's.

Price the trade explicitly: if AWS offers 2 extra points for years four and five, that is roughly 2 percent of two years of spend against surrendering a full negotiation cycle plus every off-ramp you would otherwise renegotiate.

On a $10M annual commit, the extra points are worth about $400K and the foregone cycle is typically worth 4 to 7 points on the whole base.

Ramp shape is the second fight. AWS prefers a flat commit because it front-loads revenue certainty and puts the shortfall risk on you in year one. Push for a back-loaded ramp that tracks your actual migration curve, with year-one commit set at or below current run-rate spend.

A back-loaded ramp on a three-year deal frequently beats a flat five-year deal on total discounted value, because you stop paying for capacity you have not built yet.

Timing decides composition, not just size. Offer mix (credits versus discount) shifts quarter to quarter and hardens near AWS fiscal year end, when the deal desk has budget for credits it will not carry into January.

Run your countersignature window into the last three weeks of a quarter and ask which lever the desk has left.

The counter to the five-year pitch is not refusal, it is restructuring: accept three years with a unilateral option to extend two more at pre-agreed discount floors and pre-agreed price protection scope.

You keep the option value, AWS books the pipeline, and if the market moves against you in 2029 you simply do not exercise. Frame it the way you would a multi-year enterprise agreement with an extension right rather than a term commitment, and most deal desks will take it.

The pattern that recurs is not about how hard buyers push on the percentage. It is about which document they push on.

Accounts that accepted the first structured offer forfeited an estimated 15 to 25 percent of total contract value over the term, and the loss almost never showed up as a lower headline discount.

It showed up as gross spend that failed to retire commit, as a price protection clause scoped to on-demand list rather than to the services actually consumed, and as a term extension traded away for a point that a quarter-end close would have delivered anyway.

The published UK public sector Private Pricing Addendum (DVLA, discount term 1 April 2024 to 31 March 2027) is the cleanest public artifact of this: it defines Commitment-Eligible Fees to exclude taxes, exclude fees settled from Available Balance, and count net of applicable discounts.

And it carries a Professional Services discount expressed as a 0 to 10 percent range rather than a fixed figure.

A range in a signed contract is a lane the vendor expects to be negotiated, and most buyers walk past it.

The second recurring conflict is offer composition. EDP and PPA offers arrive as a mix of credits and discounts, and that mix changes quarter to quarter depending on what the account team is being compensated on.

A credit-heavy package inflates the apparent value of year one and evaporates by year three. Treat a credit as worth roughly its face value only if it retires against fees you were already going to incur in that same period.

Separately, the 25 percent Marketplace contribution cap toward commit is its own mechanism, not part of the discount, and every dollar of third-party software above that ceiling is spend that does not count.

Same discipline applies across vendors: the discount is on the order form, the money is in the clauses.

What is the maximum discount AWS will give on an EDP or PPA?

The observed range is 5 to 20 percent, stepped by commitment thresholds, and it does not go above 20 percent. A $1M to $3M annual commit typically lands at 8 to 12 percent and a $5M to $10M commit at 12 to 16 percent.

Public sources quoting 25 percent or more are conflating the base discount with service-specific pricing layers or with credits. Model your business case against 20 percent as an absolute ceiling.

Is an AWS PPA different from an EDP?

Commercially, no. AWS deprecated the EDP label and now presents essentially all commercial offers as a Private Pricing Addendum or Private Pricing Term Agreement. The commitment structure, discount tiers and Marketplace eligibility rules are unchanged.

If your prior contract was called an EDP and your renewal paper says PPA, reuse your prior redlines but re-verify the commit-eligibility definition and Marketplace scope, both of which have moved.

How much of my AWS commit can I retire through AWS Marketplace?

The standard cap is roughly 25 percent of annual commit, and this is a separate mechanism from your discount rate, not part of it. It is negotiable upward to 30 or 35 percent in most enterprise deals.

A May 2025 policy change narrowed eligibility to SaaS products fully deployed on AWS, so validate your specific vendor list against current eligibility before you build any of that spend into your retirement model.

Should I sign a three-year or five-year AWS commitment?

One to three years is the buyer-side preference because it preserves your renewal cycle, which is the only recurring leverage event you have.

AWS will offer additional discount points for a five-year term, but given the discount ceiling of 20 percent, those points are worth a few percent while the term costs you two extra years of locked pricing in a market where AWS has said it will be capacity-constrained through 2027.

Take three years with a pre-agreed extension option instead.

Why are credits worth less than discount points in an AWS offer?

EDP and PPA offers arrive as a mix of credits and discounts, and the composition changes quarter to quarter. Credits expire, do not reduce your run-rate cost base, often carry service restrictions, and do not lower the baseline your next renewal is priced against.

A discount point compounds across every dollar of eligible spend for the full term. Convert every offer variant to a single blended three-year effective rate before you compare them.

What is the most commonly missed clause in an AWS PPA?

The Commitment-Eligible Fees definition. Published AWS addendum language counts fees incurred under Eligible Accounts but excludes fees paid from Available Balance, excludes taxes, and measures net of applicable discounts.

That last exclusion means a higher discount mechanically slows your retirement against the commit. Negotiate account scope to include affiliates and future acquisitions, and require a quarterly reconciliation statement with a defined dispute window.

Does my AWS discount survive an acquisition or divestiture?

By default, usually not on favorable terms. Standard assignment and change of control language lets AWS treat a change of control as a trigger to renegotiate, and there is typically no automatic commit reduction on divestiture.

AWS will generally concede assignment to an acquiring entity with notice and will discuss a defined divestiture carve-out, but it will resist unilateral termination for convenience without a shortfall payment. Price that shortfall exposure explicitly rather than leaving the clause open.