Finding the AWS Commit Tier Break: Why $1.4M to $1.5M Pays and $2.5M to $3M Does Not

AWS discount bands are step functions, not slopes, which means there is exactly one commit number on your spend curve worth paying for and every dollar above it is unpaid inventory. This is the method for locating that edge, proving it to your account team, and refusing the round number they will put in front of you instead.

The Step Function Nobody Draws for You

The single most useful thing to understand about AWS private pricing is that the discount curve has no slope. It has stairs. Inside a band, every incremental dollar you commit earns you exactly zero additional discount percentage while adding a dollar of shortfall exposure, and then at one specific number the rate jumps by two to four points. Look at what the market actually reports and the shape becomes obvious. One credible band map puts $500K to $2M at 5 to 8 percent and $2M to $5M at 8 to 12 percent, which places the seam at $2M. Another, drawn from a different deal set, runs $1M to $3M at 8 to 12 percent and $3M to $10M at 10 to 16 percent, which places the seam at $3M. Both are honest readings of real agreements. They disagree because there is no published grid. The seams are artifacts of how a given account team was compensated, what their quota gap looked like that quarter, and whether your logo was a competitive win against Azure or GCP. That is the leverage point: a threshold that exists only because someone inside AWS decided where to draw it is a threshold you can argue about, and one you can sometimes move down toward your number rather than moving your number up toward it. Our read on where the discount bands actually sit by spend tier is the reference; this article is the method for finding the edge of whichever band applies to you. The practical consequence is narrow and non-negotiable: the only defensible commit sits a few percent above a seam. Mid-band is the worst position on the entire curve, because you have bought the shortfall risk of a larger deal and the discount rate of a smaller one.

A threshold that exists only because someone inside AWS decided where to draw it is a threshold you can argue about.

Working the Math: $1.4M at 8% Beats $2.5M at 10%

Run the arithmetic and the headline stops sounding contrarian. Assume your realistic three-year spend curve supports about $1.45M of annual consumption. Commit at $1.45M, land the 8 percent step, and you collect roughly $116K of discount against spend you were always going to incur. Nothing is stranded. Now take the deal AWS actually proposes. Their opening commitment, in the deals we have advised, sits 15 to 30 percent above the customer's realistic curve, and the round number they will write down is $2.5M or $3M because round numbers close faster than accurate ones. At $2.6M and 10 percent, the nominal discount reads $260K, which is why the slide looks compelling. But you will consume $1.45M to perhaps $1.7M, leaving $700K to $900K of committed dollars that convert at term end into a shortfall true-up paid at 100 cents on the dollar. Your $260K of paper discount is now funding a $900K invoice for capacity you never ran. Net position: materially worse than the smaller commit, and worse in cash, not just in theory.

Three things fall out of that table and each is a negotiation instruction. First, the mid-band low commit is a genuine loss, not a safe hedge: you left two points on the table and discounted only two thirds of your spend. That is the argument for pushing up to the seam, and it is the argument your account team will happily agree with. Second, the gap between $1.45M and $1.55M is only $100K of commit but it flips the deal from 8 percent net to under 2 percent net, which tells you the tolerance on your forecast has to be tighter than the tolerance on your negotiating posture. Third, and this is where AWS's own mechanics work against them, the commitment is measured against gross spend before the discount is applied, so a commit set at your realistic consumption number is not aggressive, it is arithmetic. Do not let anyone reframe it as sandbagging. Before you name any figure, price the alternatives that buy rate without buying spend: a three-year term instead of one moves a $1M-class commit from roughly 10 percent toward 15 percent at the same dollar level, and service-level adders sit on top of the cross-service rate. Our analysis of the discount premium AWS pays for three years instead of one is the cheaper lever in almost every case. First move: rebuild the spend curve bottom-up, then compute net effective discount after shortfall for every commit number AWS proposes, and put that column on the screen in the room.

How to Locate the Seam Before You Name a Number

You cannot find a step function by asking for one price. AWS will quote a single commit with a single percentage, and that quote is engineered to look like the only offer on the table. The buyer-side counter is a grid request, in writing, in one document: quote the cross-service discount at six commit levels spanning your forecast band, three below and three above. For a customer whose realistic three-year spend curve sits near $1.5M, that means $1.2M, $1.4M, $1.6M, $2.0M, $2.4M, and $3.0M on a single sheet. The moment those six numbers sit next to each other, the flat runs become obvious. You will typically see something like 7 percent holding across $1.2M and $1.4M, then a step at $1.5M or $1.6M, then another flat run all the way to $2.4M where you are buying $800K of incremental commit for zero incremental rate. That flat run is the whole argument. It is invisible in a one-line quote and undeniable in a grid.

Two controls keep the test honest. First, hold term constant across all six rows. Term is a separate lever worth roughly five points on its own (published market reads put a one-year deal near 10 percent and a three-year near 15 percent at the same spend level), and if AWS quotes your low commits at one year and your high commits at three, you are reading a term premium and calling it a tier break. Run the grid at three years, then run it again at one year if you want the term premium priced separately. Second, require the measurement basis on the same sheet. AWS defaults to measuring consumption against gross, pre-discount spend, and a grid that silently switches basis between rows is not comparable. Ask the question explicitly: is each of these six commit numbers satisfied by gross consumption or by net invoiced dollars?

Expect resistance, and read it correctly. An account team that will only quote one number is not following policy, it is protecting a seam that would make its proposal look like an upsell. That refusal is itself the finding. Escalate it: put the six-row request in an email to the account executive and copy the deal desk contact, with a deadline tied to your internal approval calendar. Teams that hold a defensible position produce the grid within a week.

An account team that will only quote one number is protecting a seam that would embarrass its own proposal.

What AWS Does When You Refuse the Round Number

The response sequence is predictable enough to script, and knowing the order removes most of its force. Move one is the governance claim: tiers are fixed, internally approved, not something the account team controls. Treat this as a negotiating posture rather than a fact. Above-band exceptions are documented for accounts that represent a competitive win against Azure or Google Cloud, that carry unusual public reference value, or that will concede on AWS-preferred services such as Bedrock, SageMaker, and AWS-published Marketplace software. If your account has any of those attributes, say so out loud and ask which exception path applies to you.

Move two is the arithmetic switch. When percentages stop helping, AWS reframes in absolute dollars: "at $2.5M you capture another $140K of discount." That framing is true and irrelevant, because it prices the discount without pricing the commit that buys it. Your reply is a two-line comparison: the incremental discount dollars against the incremental committed dollars, plus the shortfall exposure on the delta. If $1M of extra commit yields $140K of extra discount and you are only 80 percent confident of consuming it, the expected value is negative before you count the ratchet that carries the inflated floor into year two.

Move three is the growth story, which recasts an overcommit as belief in your own roadmap. This is the most effective one because it makes prudence look like small thinking in front of your CTO. Neutralize it by moving the growth upside into a ramp band with a floor and a ceiling rather than into the year-one commit. Move four is deal desk escalation: the inflated commit survives, dressed with credits, training funds, or a Marketplace allowance. Price those sweeteners at your realized value, not face value, and net them against the shortfall risk you are absorbing.

In advised deals, AWS opening commitments sit 15 to 30 percent above the customer's realistic spend curve. Treat the round number as an ask, not a requirement, and check your position against where the bands actually sit by spend tier. A strong outcome: commit at the seam, take the same rate AWS attached to the number 40 percent higher, and hold the difference as unbought inventory.

Substitutions That Buy the Next Band Without Buying the Spend

Sometimes the seam sits genuinely above your curve. You model honestly, the next band starts at $2M, and your realistic three-year run rate tops out at $1.6M. The wrong response is to buy the gap. The right response is to pay AWS in a currency that costs you less than a $400K unfunded commit, because AWS's field team is compensated on committed contract value and multi-year lock, not exclusively on the annual number. That gives you four substitutions worth pricing. Term is the largest: market reporting on comparable deals shows roughly 15 percent on a three-year term against roughly 10 percent on a one-year term at the same commit level, which means term alone can outperform a $1M commit increase on the same spend curve. Read the three-year premium against the one-year baseline before you concede term, because you should be selling it, not surrendering it. Second, service-mix concessions: AWS will pay above-band for measurable adoption of the services it is pushing, currently Bedrock, SageMaker, and AWS-published Marketplace software, and those adders are separately negotiable from the cross-service rate, as the service-level adder benchmarks show. Third, reference value: a named logo, a case study, a re:Invent session, an analyst reference call. In my experience these are traded away for free in roughly half of deals I see, and they carry real internal weight for an account team defending a discount exception. Fourth, Marketplace spend retirement, which AWS can negotiate to count toward the commit. That one is the cleanest of all: it converts third-party software spend you already have into commit coverage without adding a dollar of new AWS consumption. Price each of these on paper with a number attached. Concede them one at a time, in exchange for a stated basis point movement, and never bundle two into a single ask.

Clauses That Make an At-Seam Commit Safe to Sign

An at-seam commit is only safe if the paper absorbs forecast error, because the seam is by definition the point where you have no cushion. Four protections carry the weight. Confirm gross measurement in writing: AWS's default is to measure consumption before the discount is applied, which is what you want, but I have seen redlines quietly flip it to net, and at a 12 percent discount that turns a $1.5M commit into $1.7M of required consumption. Demand carry-forward of unused commit into the following period, which turns a timing miss into a deferral rather than a payment. Demand a shortfall tolerance band of 5 to 10 percent with a 30 to 90 day cure period, so a soft quarter does not become an immediate true-up invoice. And insist on a ramp floor and ceiling rather than a fixed annual step, with the discount applying anywhere between them, which is the structure AWS has internal authority to approve but will not offer unprompted. Understand the two mechanics working against you. The ratchet means no subsequent year commit can fall below the prior year, so every uplift you accept is permanent. And the exit arithmetic is brutal: a mid-term termination exposes the full unfunded remaining commit, so a five-year deal signed 20 percent high is not a 20 percent problem, it is a multi-year liability you cannot unwind. That asymmetry is exactly why the at-seam number, protected by these clauses, beats a round number protected by nothing.

A five-year deal signed 20 percent high is not a 20 percent problem, it is a multi-year liability you cannot unwind.

Sequence matters. Get gross measurement confirmed first, in writing, before any commit number is discussed, because it changes what every number on the table actually means. Then table carry-forward and the tolerance band together as a single non-negotiable package, framed as the condition on which you can commit precisely rather than conservatively. AWS will trade clause flexibility for term length and forecast transparency more readily than it will trade basis points, so spend that willingness on the protections rather than on chasing another point of headline rate. If the account team refuses carry-forward outright, that is a signal they expect you to undershoot, and you should recheck your own model before signing.

Your First Three Moves

Move one happens before you take a single call: build the 24-month realistic consumption curve, net of Savings Plans and Reserved Instance coverage, because the EDP percentage lands on what is left after those commitment discounts apply. If you have not run that math, read the working of your real AWS discount after Savings Plans first, because a headline band rate applied to a shrunken base is the single most common way buyers over-buy commit. Curve first, number second. Every buyer who reverses that order signs the round number.

Move two is a written request, dated, for a multi-level discount grid: rates at the band below your curve, at your curve, and at the band above it. Ask in email so it is answerable in writing and traceable through the quarter. AWS will respond with one quote at the number they want, so restate the request and let the silence sit. This is where you assert a benchmark without exposing where it came from, using the method in proving a benchmark without naming the source.

Move three: name your commit at the first seam at or below your curve and hold it. Then trade term, service mix, and Marketplace routing for the rest of the rate rather than adding commit dollars. Expect the account team to counter with a ramp, an above-curve Year 1, or a strategic-customer framing that requires 20 to 30 percent more commit for one or two points. Decline. Market experience says the seam moves for AWS-favoured services long before it moves for raw dollars.

A strong outcome, in numbers: commit within 5 percent of the seam, band rate in the upper half of the published range for that band, annual uplift capped in writing, unused commit carried forward, and shortfall tolerance at 10 percent with a 60 to 90 day cure. If you get four of five, sign. If you get two, the seam was wrong or the timing was.

Related advisory: the full topic guide.

Frequently asked questions

Is there a published AWS discount tier table I can just look up?

No. AWS does not publish EDP or Private Pricing Agreement discount rates, and account teams are instructed not to share benchmark data. The band ranges circulating in the market are reconstructions from deal observation, which is why two credible sources place the seams at different dollar amounts. Treat them as a starting hypothesis and force AWS to quote across multiple commit levels to find where your seam actually sits.

How much extra commit is usually needed to cross a band edge?

It varies by deal, but the gap is often small in percentage terms. Advisors have seen increments in the range of $80K to $100K of additional annual spend unlock the next tier. That is worth paying for if the incremental dollars are genuinely consumable. It is not worth paying for if crossing the seam requires inflating commit by 20 percent or more above your realistic curve.

What happens if I commit above what I actually consume?

AWS's standard shortfall clause requires you to pay the difference between the committed and actual amount at the end of the measurement period, at full value. There is no partial credit. That is why an inflated commit at a higher band rate frequently produces a worse net outcome than a smaller commit at a lower rate, and why shortfall tolerance and carry-forward are the two clauses worth spending negotiation capital on.

Can AWS give a discount above the band my commit falls into?

Yes, though not on request alone. Above-band pricing is approved where the deal represents a competitive displacement of Azure or Google Cloud, where the account carries unusual public reference value, or where the buyer concedes on AWS-favored services and Marketplace purchasing. Expect to pay for it in non-cash currency, and price that currency before you hand it over.

Should term length or commit size be the lever I push first?

Test term first, because it costs no shortfall exposure. Moving from a one-year to a three-year agreement at the same commit level has been observed to lift the rate from roughly 10 percent to roughly 15 percent. If term gets you to the same effective rate as the next commit band, take the term and leave the commit at your realistic curve.

Is my commit measured on gross spend or net of the discount?

AWS's default is to measure against gross spend before the discount is applied, which favors the buyer because your consumption retires the commit faster. Confirm it explicitly in the agreement. If a draft measures net, a $2M commit at a 15 percent discount effectively requires about $2.35M of gross consumption to satisfy, which quietly moves your break-even point.