AWS will not sign a clean termination for convenience, but it will sign a post-Month-24 walk at 50 percent of one year's commit instead of the full remaining term
The AWS Customer Agreement already grants termination for convenience in Section 5.2(a); the PPA addendum is drafted specifically to override it, leaving the default exit at unused remaining commit in full, a figure that reaches $33M on a Year-2 walk from a five-year $10M ramping deal. The negotiable ground is not whether you can walk but what walking costs and what triggers it, and the realistic landing zone is a 50 percent one-year fee, a 15 to 40 percent step-down, and a 30 to 90 day cure window. If you do not draft these before signature, they cannot be added later.
Prepared by Redress Compliance · September 8, 2026 · AWS private pricing advisory. EDP and PPA negotiation and redline engagements, 2024 to 2026.
Executive summary
The right you are asking for already exists in AWS public paper, and the addendum exists to take it away.
Section 5.2(a) of the AWS Customer Agreement lets a customer terminate for any reason on notice and account closure, so every PPA negotiation over termination is a negotiation about how far the addendum overrides a right AWS already publishes.
The default exit price is not a penalty percentage, it is the whole unfunded balance: $33M on a Year-2 walk from a five-year $10M ramping commit, with documented exit penalties exceeding 75 percent of remaining unused commit.
AWS frames the commit as a guaranteed minimum spend rather than a usage target, which is why shortfall is invoiced rather than forgiven and why the exit math is punitive by design.
The clause AWS actually signs is a capped walk: termination after Month 24 on 180 days' notice at 50 percent of one year's minimum commitment, not 50 percent of the remaining term.
That single word change converts a $33M theoretical exposure into a single-digit-millions cash number on a $10M-per-year deal, and it is the highest-value redline on the page.
Where AWS refuses a walk right outright, the fallback stack is a 25 to 40 percent step-down on enumerated triggers, a narrower 15 to 20 percent version on 90 days' notice, and a 30 to 90 day cure window that together buy 15 or more months of flexibility.
Expect AWS to counter by lengthening notice periods and offering price for term instead, and expect to trade discount points, not commit dollars, for the exit.
Three off-ramps AWS leaves silent by default decide whether you ever get to use the clause: force majeure drafted to natural disasters only, an assignment clause that triggers on change of control.
And governance terms that appear in roughly seven out of ten PPAs when requested explicitly and never once when they are not.
Silence in the draft is not neutrality; it is AWS's position.
How the termination stack actually works: MSA right, addendum override, RI precedent
The first thing to understand is that AWS is not defending a position it holds by default. The public AWS Customer Agreement, Section 5.2(a), already gives you termination for convenience: close the accounts, give notice, walk. AWS gets the same right on 30 days.
That right is drafted away by the Private Pricing Agreement addendum, which is the only one of the three documents (Customer Agreement, Service Terms, PPA addendum) where the money lives.
So when your AWS account team says "commitments are non-cancellable," what they mean is that AWS wrote a paragraph specifically to remove a right you already had.
That framing matters at the table, because it converts the conversation from "please grant us an unusual concession" to "you are asking us to give up a published contractual right.
And we want to be paid for it." Expect AWS counsel to reach for the EC2 Reserved Instance precedent, which states that RI pricing is owed for the full selected term even if the Agreement is terminated.
It is a real clause and a fair analogy for a single reservation. It is a weak analogy for a five-year, organization-wide, ramping spend floor, and you should say so on the record: an RI is a discrete capacity purchase you elected, while a PPA commit is a forecast AWS helped build.
The economics of getting this wrong are not subtle. On a five-year $10M ramping deal, walking at the end of Year 2 leaves roughly $33M of cumulative unfunded commit exposed, and market benchmarks put exit penalties above 75 percent of remaining unused commit where nothing was drafted.
The negotiated landing zone is an order of magnitude cheaper. Treat this alongside the broader AWS PPA clause redlines rather than as a standalone ask.
| Exit position | Trigger | Notice | Cash cost on 5yr $10M ramp (Year-2 walk) | AWS concession rate |
|---|---|---|---|---|
| MSA Section 5.2(a) convenience right | None; any reason | Account closure | $0, but overridden by the addendum | Not available once PPA signs |
| Default addendum (no redline) | None; you simply owe the balance | N/A | ~$33M unfunded, penalties above 75% of unused commit | 100% (this is the paper you are handed) |
| Negotiated capped walk | Post-Month-24, any reason | 180 days | 50% of one year's commit, roughly $4M to $6M | Uncommon; requires real concession |
| Step-down only | Enumerated events (divestiture, BU closure, AWS service issues) | 90 days | $0 cash; commit cut 15 to 40%, discount may reset | Most common landing point |
Read the table from the bottom up, not the top down. AWS will offer you row four and call it flexibility, because a step-down costs AWS nothing in cash and lets it re-open the discount tier to claw back margin.
Row three is the only line that gives you a genuine off-ramp, and the price of it, 50 percent of a single year rather than the full remaining term, is the entire negotiation compressed into one number.
The tactical implication: never bid for row three by itself. Ask for both, in one redline, and let AWS give you the step-down as the "win" while you hold the capped walk as the item you will trade term length or a higher commitment tier for.
Buyers who lead with the walk right alone get a flat refusal and no fallback on the page.
The five off-ramps worth drafting, and what each one is worth
AWS will sign a list. AWS will not sign a standard.
Every hour spent arguing for "materially adverse business conditions" or "commercially reasonable business need" is an hour AWS legal spends explaining that such language is unenforceable and unauditable, and they will win that argument because they are largely right. Enumerate instead.
Five triggers are worth the redline, in this order.
First, change of control and assignment protection, which the default paper leaves genuinely unaddressed: without acquirer survival language, your PPA terms are renegotiable the day your deal closes, and the acquirer inherits a commit at worse pricing.
This is the cheapest trigger to win and the one with the highest expected value, and it is worth reading against the mechanics in the AWS PPA change of control clause before you draft.
Second, material adverse change tied to divestiture, with the commit reduced proportionally to the divested entity's trailing twelve months of AWS spend. Third, budget failure and demand collapse. Do not attempt to route this through force majeure.
AWS drafts force majeure to natural disasters and acts of war, and it will not carry a demand argument, so this trigger must be its own paragraph with objective tests: workforce reduction exceeding 20 percent, a documented multi-cloud strategy shift approved at board level.
Or a regulatory change that prohibits a workload.
Fourth, sustained service degradation, drafted against documented AWS service issues over a defined measurement window, not against your own perception of performance. Fifth, a price-increase trigger paired with a price file freeze locking public list at signature.
That freeze is never volunteered, is granted on roughly seven out of ten PPAs when asked explicitly.
And is worth 4 to 8 percent across a three-year term, which makes it the highest-yield governance clause on the page and a legitimate walk trigger if AWS raises list on a service carrying more than a stated share of your spend.
Price each trigger separately at the table. Change of control and the price file freeze should cost you nothing beyond asking.
The divestiture step-down usually lands at 15 to 20 percent on 90 days' notice, once per term, with the discount tier subject to adjustment, and pushing to the 25 to 40 percent band is a genuine negotiation you win by extending term length or accepting a higher Year-1 ramp.
The budget-failure trigger is the hardest of the five and the one to trade last.
Attach a 30 to 90 day shortfall cure window to all of them; stacked with a step-down it produces roughly fifteen months of practical runway before any true-up bites, which in most cases is more useful than the walk right you were originally chasing.
The recurring mistake is treating these as five alternative asks. They are one system: the enumerated triggers decide when you may act, the step-down percentage decides how much relief you get without cash, and the capped walk decides what the worst case costs.
Drop any one and the other two lose most of their value.
Protect your AWS EDP commit with 6 flexibility clauses
Six flexibility clauses protect an AWS EDP commit: rollover, carryforward, over commit caps, under commit relief, and clean exit ramps.
Get the white paper →Why AWS trades notice periods instead of walk rights, and how to make that work for you
Watch what the account team actually does when you table a termination right. It does not say no. It says yes, with a notice period long enough that no rational CFO would ever pull the trigger: 180 days on a twelve-month deal, which is to say the right expires before it becomes useful.
The alternative deflection is a pricing fork, better economics for a longer term or worse economics for a shorter one, which reframes your risk question as a discount question. Both moves come from the same page of the playbook.
The objective is never to withhold the right, because withholding it invites escalation and legal involvement. The objective is to price the right into irrelevance so it can be conceded on paper and never exercised in practice.
Read that as an opening rather than a wall. Notice length costs almost nothing to a buyer who plans migrations in quarters and everything to a buyer who is improvising.
If your platform team can scope a workload exit, stand up the target environment, and cut over inside two quarters, then 180 days of notice is a calendar entry, not a cost. Push it to 270 days and you are still inside the planning horizon most large enterprises already use for infrastructure moves.
What you are buying with that concession is the number that actually matters: the fee cap. Trading notice length for a walk fee capped at 50 percent of one year's commitment, rather than the full unfunded remainder, is a present-value trade the buyer wins by a wide margin.
On a five-year ramping deal, the difference between 50 percent of one annual minimum and cumulative unfunded commit is the difference between a manageable write-off and a nine-figure conversation with your audit committee.
The asymmetry in AWS's own incentives is stronger than the account team will ever admit. A walk right exercised in Year 3 costs AWS a renewal it was going to have to compete for anyway, against Azure, against GCP, against your own repatriation business case.
A full-balance clawback, by contrast, is a receivable AWS has to chase, possibly litigate, and book against a customer relationship it wants to keep. Legal departments on both sides know that the enforceable value of a clause you have to sue to collect is a fraction of its face amount.
The clause earns its keep before it is ever tested, by preventing churn through fear. That is precisely why AWS will trade its exercise conditions freely and its fee cap grudgingly.
The real source of the resistance is internal, and understanding it tells you which currency to bring. The commitment number underwrites the discount band, and the discount bands are narrow: 5 to 20 percent, stepped by commitment threshold, never above 20.
That number also underwrites the account team's quota and the forecast the deal desk has already socialised internally.
A clause that lets the number shrink mid-term is therefore not a legal concession, it is a revenue concession, and it has to be funded out of discount points rather than commit dollars. So stop asking for the walk right for free.
Ask for it against a slightly higher commitment tier, or accept a discount that sits a point or two below the top of your band, or take a larger share of the offer in credits, which shift quarter to quarter anyway. Buyers who bring the right currency get the clause.
Buyers who treat it as a pure legal ask get the 180-day notice period and nothing else.
Sequencing decides the outcome more than drafting skill does. Raise termination in the same conversation as term length and ramp shape, while the commercial envelope is still open and the account team still has room to reshape the deal in your favour.
At that stage the walk right is a variable inside a live model. Raise it after the commit number, the ramp, and the discount are agreed, and it becomes a one-way concession with no funding source, which is why late-cycle termination asks fail regardless of how well the language is written.
The same rule governs the surrounding governance package covered in our AWS PPA clause redlines analysis: these provisions are never volunteered and cannot be retrofitted after signature.
Drafting the clause: language that survives AWS legal review
The clause has to be specific enough that AWS legal cannot reinterpret it at true-up time and narrow enough that the deal desk can price it. Four provisions do the work: the capped walk, the step-down, the cure window, and credit carry-forward. Draft all four, expect to land two or three.
Note the traps in the right-hand column, because that is where AWS legal recovers what the account team conceded.
| Provision | Operative language to table | Where AWS legal claws it back |
|---|---|---|
| Capped walk right | Customer may terminate this Addendum after Month 24 on 180 days' prior written notice, subject to a termination fee equal to 50 percent of the annual minimum commitment then in effect for one year, and expressly not the aggregate remaining commitment for the balance of the term. | Survival language drafted on the EC2 Reserved Instance model, under which amounts are owed in full "even if the Agreement is terminated," which reinstates the full balance behind your cap. |
| Step-down right | Customer may reduce the annual minimum commitment by up to 20 percent on 90 days' written notice, once per term, on any of: divestiture, workforce reduction exceeding 20 percent, closure of a named business unit, or documented sustained service degradation. | "The discount percentage may be adjusted to reflect the reduced commitment," which converts a 20 percent relief into a discount reset and can leave you worse off in net spend. |
| Shortfall cure window | Any shortfall against the annual minimum is subject to a 90 day cure period during which incremental consumption, Marketplace spend, and prepaid commitments count toward cure, before any true-up is invoiced. | Cure defined to exclude Marketplace, which matters because Marketplace contribution is separately capped at 25 percent and is a distinct mechanism from the discount. |
| Credit carry-forward | Any unmet portion of the annual minimum converts to AWS service credits usable in the following 12 months rather than a cash true-up. | Credits scoped to a service list that excludes where your growth actually is, or expiry set to the term end so Year 5 credits are worthless. |
| Scope of termination | Termination of this Addendum shall not require closure of the AWS account or termination of the AWS Customer Agreement. | Silence here lets AWS argue the MSA Section 5.2(a) mechanic applies, meaning your only clean exit is closing every account. |
The table's real lesson sits in the last row. The AWS Customer Agreement already grants termination for convenience; the PPA addendum exists partly to override it, and the override works by tying any exit to account closure.
If your clause does not separate terminating the pricing addendum from terminating the relationship, you have drafted a right you cannot use, because no enterprise closes production accounts to escape a discount schedule.
The second lesson is that a step-down with a discount reset attached is not flexibility, it is a rebate reduction dressed as relief. Model both provisions in net spend, not in commitment relief, before you agree to either.
Where the discount is the currency, as it is in every one of these trades, price protection and reservation-product carve-outs interact with the walk fee, which is why the price protection gap on reservation products belongs in the same redline pass rather than a later one.
What the deals show: patterns across PPA redline engagements
Price file freezes, change of control protection, and audit reciprocity were never volunteered in a default AWS draft but landed on roughly seven of ten PPAs when the buyer requested them explicitly and in writing.
Where AWS refused a walk right outright, the traded alternative was a shorter term with a Year-3 re-price, and the buyer reached the same total spend as the five-year deal at roughly 15 percent uplift on the annual commit.
The pattern across redline engagements is consistent enough to plan around. AWS almost never opens with any of the clauses that make a commit walkable, and it almost never refuses to discuss them once a buyer names them specifically.
What gets granted is asymmetric: step-downs land far more often than walk rights. A 15 to 20 percent reduction on 90 days' notice, one use per term, tied to enumerated triggers, is close to routine when the trigger list is narrow and evidence-based.
The 25 to 40 percent step-down appears where the buyer has a live divestiture or a documented service failure history. A clean post-Month-24 walk at 50 percent of one year's commit sits at the top of the achievable range and requires the buyer to spend real currency to get it.
When a walk right is refused outright, the substitute AWS offers is term compression rather than fee reduction. AWS would rather sell three years at a higher annual commit than five years with an exit. That trade is worth taking.
In documented cases the compressed term reached the same aggregate spend at roughly 15 percent higher annual commit, with materially less unfunded balance exposed at any point in the term.
A three-year deal with a Year-3 re-price and no walk right carries less cash risk than a five-year deal with a walk right you have to litigate.
Two mechanical patterns matter because they cost buyers leverage. First, AWS has standardized on PPA as the label for all negotiated private pricing, and buyers still drafting to legacy EDP definitions invite scope arguments over which services and which addendum the clause attaches to.
Draft to PPA and cross-reference the addendum by exhibit number.
Second, the trigger language in a step-down or walk clause is worthless if the commit itself is measured ambiguously, so the definition work covered in our commit measurement and audit clause guidance and the broader AWS PPA clause redline set has to be settled before the termination fee is priced.
The same applies to assignment and change of control, where the default draft leaves an acquirer exposed to renegotiation, and to the confidentiality terms that determine whether you can benchmark your own price file at renewal.
- Percentile standing for your exact deal size and industry, from real closed transactions
- Scenario simulation before the call: test alternative terms and see the financial impact of each
- A negotiation playbook, talking points, and a two page executive brief on day one
Your first five moves
- Model the unfunded balance in cash before the first pricing call, so the Year-2 and Year-3 exposure (on a five-year $10M ramping structure that reaches roughly $33M at end of Year 2) sits on the table next to the discount percentage rather than surfacing in legal review three weeks before signature.
- Ask for the capped walk at 50 percent of one year's commit, and when AWS pushes back, concede notice length rather than fee level: 180 days or even 270 days costs you nothing operationally, while every point of fee movement is real cash.
- Enumerate five named triggers instead of requesting broad convenience, because AWS legal will reject anything resembling Section 5.2(a) in the addendum but will engage on divestiture, business unit closure, workforce reduction above 20 percent, documented sustained service degradation, and a public price increase above a stated threshold.
- Demand the 30 to 90 day shortfall cure and credit carry-forward as the named fallback, treating them as non-optional because they cannot be added post-signature, and stacked with a step-down they buy 15 or more months of effective runway even if the walk right is refused.
- Hold two discount points in reserve as the currency for the clause, and never pay for flexibility with commit reduction, since raising the commit to buy an exit right inverts the entire trade and AWS will take that deal every time.
The sequencing is the whole point.
Buyers who lead with the discount conversation and raise termination later find that AWS has already priced the deal on the assumption of a hard commit, and reopening the exit then reads as a late-stage escalation the account team can route to a "that structure is not available" response.
Buyers who put the Year-2 cash exposure on the table in the first pricing call reframe the negotiation: the discount is no longer the only variable, and the account team now has to win on two dimensions.
Remember what the discount band actually is. AWS EDP and PPA discounts run 5 to 20 percent, stepped by commitment threshold, and never above 20 percent.
The offer is a mix of credits and discounts whose composition shifts quarter to quarter, and the 25 percent Marketplace contribution cap toward commit is a separate mechanism entirely.
That ceiling is your friend here: because the discount is bounded, the marginal value of the last point is small, which is precisely why two points are the right currency for a walk right worth eight figures in avoided exposure.
Frequently asked questions
Does the AWS Customer Agreement already let me terminate for convenience?
Yes. Section 5.2(a) of the public AWS Customer Agreement allows the customer to terminate for any reason by giving notice and closing the account for services with an account closing mechanism, and allows AWS to terminate on at least 30 days' notice.
The point of the PPA addendum is to override that right for the committed spend, which is why AWS argues so hard for commitment survival language. Read the addendum's survival clause, not the MSA, to find your actual position.
What does it cost to exit an AWS PPA if I have no negotiated termination right?
The default is payment of the unused remaining commit at full value.
On a five-year $10M ramping commit, a customer walking at the end of Year 2 faces roughly $33M in cumulative unfunded commit plus any contractual penalties, and documented exit penalties have exceeded 75 percent of remaining unused commit. That is the number to model before you accept a term length.
What termination language will AWS realistically sign?
The most useful documented precedent is a right to terminate after Month 24 with 180 days' notice, subject to a fee equal to 50 percent of the remaining annual minimum commitment for one year rather than for the remaining term. The one-year cap is the whole point of the clause.
AWS will resist it, will counter with longer notice, and is more likely to concede it when it is raised alongside term length rather than as a late legal ask.
Will AWS agree to a commit step-down instead of a full exit?
More often, yes. AWS will sometimes agree to reduce annual commit by 25 to 40 percent on specific triggers such as an M&A divestiture, a business unit closure, or demonstrated AWS service issues, and a narrower version permitting 15 to 20 percent on 90 days' notice appears more frequently.
Expect one exercise per term and expect AWS to propose adjusting the discount percentage to reflect the smaller commitment. Negotiate the step-down as the primary mitigation and treat the walk right as the stretch ask.
Does AWS force majeure cover a budget cut or a collapse in demand?
No. The AWS default force majeure language is drafted around natural disasters and acts of war and is rarely applicable to changes in cloud consumption. Budget failure, demand collapse, and strategy changes have to be written in as named triggers for a step-down or termination right.
If you rely on force majeure to carry a commercial argument, you have no off-ramp.
Can termination flexibility be added after the PPA is signed?
Practically, no. Shortfall cure periods of 30 to 90 days, step-down and adjustment rights, credit carry-forward, and mid-term ramp adjustments all have to be secured at drafting time; once the addendum is executed AWS has no commercial reason to reopen it until renewal.
Stacked, a cure period plus adjacent adjustment rights can deliver 15 or more months of effective flexibility, which is why these clauses belong in the first redline pass.
What happens to the PPA if my company is acquired?
The default AWS PPA carries a standard assignment clause that triggers on change of control, which means the discount and the commit are both open to question in an M&A event.
Buyer-side markups add acquirer protection so the PPA terms survive the transaction; without that language the acquirer can face renegotiation on worse terms. Treat change of control as both an assignment issue and a termination trigger, and draft it in the same pass.
