What Service-Level Discount Adders Can You Get on Top of Your AWS Cross-Service Rate?

Your cross-service EDP percentage is only the first layer of the deal; concentrated EC2, S3, RDS or egress estates support a second, service-specific schedule worth another 3 to 15 points on that spend. This page prices that second layer, names the concentration thresholds that qualify you to ask, and shows what AWS will demand in return.

The Second Layer Is a Separate Document, and That Is Your Leverage

The single most expensive mistake buyers make in an AWS renewal is treating the deal as one percentage. It is not. The cross-service EDP discount lives in one instrument; the service-specific schedule (the private pricing addendum) is a separate document, negotiated on separate evidence, signed on its own timetable. Historically the EDP was the only contractual route to a broad rate, with the PPA bolted on as an addendum to deliver service-level pricing. That separation is the leverage. It means you can close the headline band, let the account team book the win internally, and then open a second conversation on EC2, S3, DynamoDB or egress with fresh evidence and a different decision maker (the service business unit, not just the account executive). Buyers who negotiate to a single number surrender that second conversation permanently, because once the EDP is countersigned the account team has no incentive to reopen anything.

Stacking order determines what the second layer is actually worth, and here AWS has the structural advantage. The EDP discount applies to on-demand pricing after Savings Plans and Reserved Instances have already been applied; it stacks rather than replaces. So a 10 point EC2 adder does not reduce your compute bill by 10 percent. It reduces the residual on-demand-priced slice that survived your commitment coverage. If you already run 80 percent RI and Savings Plan coverage on compute, the adder is discounting a fifth of the estate. AWS knows this arithmetic better than most buyers do, which is why compute adders are the ones they concede most readily. Model the effective dollar value before you spend negotiating capital on the point count, and see how the base band behaves in our AWS private pricing discount bands by spend tier.

A ten point EC2 adder does not cut your compute bill by ten percent; it discounts only the residual on-demand slice your Savings Plans did not already cover.

Benchmark Adder Bands by Service: What the Second Layer Is Actually Worth

The bands below are what concentrated estates achieve in 2026, not what AWS opens with. AWS will open at the bottom of each range or claim the service is out of scope entirely. Bedrock sits at the top of the table for a reason that has nothing to do with your negotiating skill: AWS is buying AI adoption and is willing to fund it out of margin, which makes it the cheapest point to win and the most dangerous to over-commit against, because you will be signing a service-level minimum on consumption you cannot forecast.

Scope is the first fight. A private pricing schedule generally reaches EC2, ECS, EKS, Lambda and Fargate; S3, EBS, EFS, Glacier and FSx; RDS, DynamoDB, Redshift, ElastiCache, EMR, Athena and Kinesis; and the network layer of VPC, Direct Connect, Route 53, CloudFront and Transit Gateway. If a service you spend heavily on is not on that list, you are asking for a bespoke exception and should price the ask accordingly.

Now the envelope. EDP tiers run 5 to 25 percent on net spend, with the top band gated behind five-year terms and $50M-plus annual commitments; the service schedule adds another 5 to 15 percent on named services only. On a $10M estate that sounds like a combined 40 percent. It is not. If EC2 is 35 percent of spend at $3.5M, and Savings Plans already cover 75 percent of it, a 10 point adder touches roughly $875K of residual and returns about $88K. The same 10 points on a $2M unreserved egress and network line returns $200K. That asymmetry is the whole planning exercise: identify the concentrated, uncommitted slice, because that is where the second layer actually pays. In our experience across these renewals, buyers routinely spend their leverage on the compute adder (the loudest line on the invoice) and leave the network schedule, which is worth two to three times more in cash, entirely unasked.

Do You Actually Qualify? The Concentration Test AWS Applies

The account team underwrites the second layer against shape, not size. A $4M estate spread evenly across forty services gets nothing extra; a $900k estate where S3 is 34% of the net bill gets a schedule. That distinction matters because asking without the profile costs you something real. The rep runs the ask up to the pricing desk, the desk declines, and your next three asks in that cycle arrive with a note attached. Save the escalation for a request the desk can actually approve. The published entry bar is roughly $300,000 in annual AWS billing, about 20% year-over-year usage growth, consistent Marketplace spend, Enterprise Support enrolled across every account in the org, and a 3 to 5 year term. Treat that as the floor for a conversation, not a qualification for a good number. Above the floor, the desk applies service-level tests: internet egress above 5TB per day, total data transfer running 10 to 20% of spend once cross-AZ, NAT Gateway processing and inter-region replication are counted, or any single service exceeding roughly 25 to 30% of the net bill.

The precondition that kills most EC2 adder asks is your own commitment coverage. Industry-average Reserved Instance coverage on compute sits at 42%; best-in-class enterprises run 75 to 85%. On a $20M bill, closing that gap is worth about $4.76M a year, which is a larger number than any EC2 adder AWS will write. If you walk in at 42% coverage, the desk knows the adder would apply a few extra points to a base you have not optimized, and it will quite reasonably decline or offer the bottom of the band. Fix coverage first, then ask. The same logic governs the broader rate discussion covered in our AWS private pricing discount bands by spend tier analysis: AWS prices the estate it sees, and a bloated base reads as leverage for them, not you.

Egress: The Widest Spread and the Fight AWS Will Deny Is Winnable

Egress is where the largest single-line dollar swing sits and where AWS's stated position is flatly contradicted by observed outcomes. Reps will tell you data transfer out is a published, non-negotiable rate at $0.09/GB. That is a script, not a policy. For estates moving more than 5TB per day, negotiated rates land in the $0.04 to $0.07/GB range. Enterprises that explicitly put egress on the table save an average of $420,000 a year, and the most concentrated profiles (video platforms, high-frequency trading, cross-region replication architectures) have taken $1.2M to $2.1M off that line alone. The gap between the script and the outcome exists because the person delivering the script does not own the decision.

Your first move is to quote AWS's own tiered schedule back at them. List runs $0.09/GB for the first 10TB monthly, $0.085 for the next 40TB, $0.07 for the next 100TB, and $0.05 above 150TB. AWS has already conceded in public that egress pricing moves with volume. The only remaining question is where your curve sits and what a private schedule does to it. Build the model: 100TB per month costs $7,980 at list, so a move to a flat $0.05/GB blended rate is roughly $2,980 a month, $36,000 a year, on a modest estate. Scale that to 5TB per day (about 150TB monthly) and the arithmetic gets serious fast.

A rep who says egress is non-negotiable is telling you about their approval authority, not about AWS pricing policy.

Refuse "non-negotiable" from anyone who is not the pricing desk. Ask the rep directly whether the answer is theirs or the desk's, and ask for it in writing. That single question converts a soft decline into an escalation. Bring the volume data to the meeting: 90 days of actual GB out, broken by internet egress, cross-AZ at $0.01/GB each direction, NAT Gateway processing, and inter-region replication. Most buyers underestimate their own network line by half because it is scattered across a dozen SKUs. A strong outcome on a concentrated estate is a blended internet egress rate at $0.05/GB or below, written into the private pricing addendum with a stated volume floor rather than a discretionary target. Anything above $0.07 on a 5TB-per-day profile means the desk never actually looked at your numbers.

The Architecture Alternative That Caps What AWS Will Pay You

An egress adder is never worth more than the cost of routing around it, and the AWS account team prices to that ceiling whether or not you put the model on the table. Direct Connect is the arithmetic AWS has already run: port fees start at $0.30 per hour ($219 per month) on 1 Gbps and reach $85.00 per hour ($62,050 per month) on 400 Gbps, while egress over the circuit drops from $0.09 per GB on the internet to $0.02 per GB from US regions. That spread sets hard breakevens at roughly 3.2 TB per month on 1 Gbps, 24 TB on 10 Gbps, and about 410 TB on 100 Gbps. Below the breakeven your bypass threat is theater and AWS will treat it that way. Above it, the reseller-neutral math says you can leave, and the adder conversation changes character in one meeting. The same logic applies to CDN offload: if a meaningful share of your egress is cacheable, model the CloudFront or third-party CDN alternative and price it, because AWS will otherwise assume you have not.

Bring the costed bypass model, not the assertion. In our experience across these negotiations, a spreadsheet showing port fees, cutover cost, and a dated migration plan moves egress rates two to three cents faster than any volume argument. Then insist the same schedule prices the SKUs that quietly inflate the line: cross-AZ transfer at $0.01 per GB in each direction and $0.005 per hour for every public IPv4 address, attached or idle. Those are inside the network bucket you are discounting, and if you do not name them AWS will scope the adder to internet egress alone and leave 20 to 30 percent of your transfer spend at list.

What AWS Charges You for the Adder: Service Commits, Term, and Uplift

The second layer is sold, not granted. AWS converts every service-level adder into a service-specific minimum inside the agreement that behaves exactly like a Reserved Instance or Savings Plan: hit it and you keep the rate, miss it and you pay the shortfall. That creates a second exposure sitting underneath your aggregate commit, and it is entirely possible to clear the top-line number while defaulting on the S3 or EC2 bucket. Expect the ask list to arrive intact: a 3 to 5 year term, roughly 20 percent annual commit growth, Enterprise Support enrolled across every account, and Marketplace spend routed through AWS. The 3 to 5 points of uplift AWS attributes to a three-year term are already priced into the base rate benchmarks, so do not let the account team sell you term twice, once for the headline percentage and again for the adder. Check the headline first against published discount bands by spend tier so you know which layer is actually underpriced before you trade anything.

The counter-position is three clauses. First, cap the service-level minimum at 70 to 80 percent of trailing twelve-month usage on that service, never at forecast, because forecast is AWS's number and trailing actuals are yours. Second, negotiate a one-time reallocation right between service buckets, effective at the midpoint of the term, so an architecture change (moving RDS workloads to Aurora, shifting storage classes, re-platforming compute) does not manufacture a shortfall you never intended. Third, refuse cross-default language outright. AWS drafting frequently allows a miss on one service minimum to trigger repricing or default on the entire agreement, which converts a $2M service bucket into leverage over a $30M commit. A strong outcome looks like this: the adder lands in the benchmark band for the service, the service minimum is capped at 75 percent of trailing usage, growth is negotiated to 10 to 12 percent rather than 20, and the only remedy for a service shortfall is a true-up on that bucket alone. If AWS will not sever cross-default, the adder is not worth the exposure and you should take the broad rate and walk the second layer to the next renewal.

Clauses That Decide Whether the Adder Pays or Costs You

Most of the money in a second-layer schedule is lost in three clauses, not in the percentage. The first is the measurement basis of your commitment counter. If your $2M annual commit burns down on net (post-discount) spend, every point of adder you win makes the commit harder to retire: at a blended 15 percent, you need roughly $2.35M of gross consumption to clear a $2M net commit, and a rich EC2 or S3 adder pushes that gross requirement higher still. AWS will present net measurement as standard and it usually is; your counter is to insist that the service-level schedule apply to invoice pricing while the commitment counter runs on pre-discount usage, or, if AWS refuses, to size the commit off net consumption modeled at the post-adder rate rather than at today's effective rate. Buyers who skip this arithmetic negotiate a deeper discount and a shortfall penalty in the same signature.

The second clause is what does not count. Marketplace purchases, promotional credits, Enterprise Support fees, third-party charges, and in many drafts the service-specific spend that triggered the adder in the first place are routinely excluded from the drawdown. Get the inclusion list written affirmatively rather than accepting an exclusion list, and confirm that spend inside the service schedule still counts toward the master commit. The third clause is drift: service schedules frequently carry their own term, expiring or resetting on an anniversary that does not match the master agreement, which hands AWS a mid-term repricing event. Co-terminate them or you will renegotiate the same adder twice.

Then there is the renewal trap. A generous adder on a service you are actively migrating off sets the baseline AWS anchors to three years out, and true-forward mechanics will hold you to a commitment built on a workload that no longer exists. Price the adder against projected, not trailing, consumption, and cap the service minimum accordingly. The same discipline applies across private pricing discount bands by spend tier: the headline number is easier to win than the terms that let you keep it.

What to Do First

Work a 30 to 60 day sequence, not a discount request. Days 1 to 15: pull twelve months of Cost and Usage Report data and rank every service by net spend and by concentration ratio (that service as a share of total AWS spend). Anything above roughly 25 percent of the bill is schedule-eligible in practice; anything below 10 percent is not worth the negotiating capital. Days 15 to 30: fix Reserved Instance and Savings Plan coverage before you price a compute adder. Best-in-class estates run 75 to 85 percent coverage against an industry average of 42 percent, and on a $20M bill that gap alone is worth roughly $4.76M a year. An EC2 adder applied to an uncovered base discounts your own inefficiency and locks it into the commit.

Days 30 to 45: cost the Direct Connect and CDN bypass for egress. At $0.02/GB from US regions against $0.09/GB over the internet, breakeven lands near 3.2 TB/month on a 1 Gbps port. Table that build as a costed alternative, not a threat. Days 45 to 60: put the ask in writing. Target EC2 plus 5 to 8 points, S3 plus 3 to 5, egress at $0.05 to $0.06/GB above 5TB/day, and a service minimum capped at 75 percent of trailing usage. When the account team says the schedule does not exist, escalate past the rep to the regional private pricing desk; the document exists, and in our experience the denial is a first-round position, not a policy. Read this alongside the sibling analysis on term premium, effective rate after Savings Plans, commit tier breaks, uplift caps, and the cost of above-band discounts.

Related advisory: the full topic guide.

Frequently asked questions

Can you get a service-specific AWS discount without an EDP?

Rarely at enterprise scale. The service-level schedule is structured as an addendum to the private pricing or EDP contract, so the cross-service commitment is normally the entry ticket. Buyers below roughly $300k annual billing with under 20% growth will be pushed toward Savings Plans and Reserved Instances instead, which is often the correct answer economically.

How much extra discount can I get on EC2 specifically?

Observed additional bands run 5 to 12 points on top of the cross-service rate for concentrated compute estates. But that adder applies after Savings Plans and Reserved Instances, so on an estate already at 50 to 65 percent blended compute discount, the incremental dollar value is modest. Fix RI coverage first: moving from 42 percent to 80 percent coverage on a $20M bill is worth roughly $4.76M a year, far more than any adder.

Is AWS data transfer out really negotiable?

Yes, above volume. AWS reps frequently state that $0.09/GB internet egress is non-negotiable, and for most accounts that is true. Above roughly 5TB per day, negotiated rates of $0.04 to $0.07/GB are documented, with the largest video, trading and replication estates taking $1.2M to $2.1M off the line annually. The credible threat is architectural: a costed Direct Connect model at $0.02/GB sets the ceiling on what AWS will concede.

What does AWS ask for in return for a service-level adder?

A service-specific spend minimum inside the EDP, a three to five year term, Enterprise Support enrolment across every account, and usually 20 percent annual commit growth. That minimum creates a second shortfall exposure separate from your aggregate commit. Cap it at 70 to 80 percent of trailing twelve-month usage on that service and negotiate a reallocation right between service buckets.

Should the adder be measured against gross or net spend?

Gross, always. If your commitment burns down on post-discount spend, a deeper service-level adder means you consume your commit more slowly and increase shortfall risk. At a 15 percent discount on a $2M commit, gross measurement is worth roughly $300k of burn-down per year. Confirm the measurement basis in writing before agreeing the percentage.

Is S3 worth negotiating a separate schedule for?

It depends on concentration. Some advisers argue storage is the hardest line to discount because it is already cheap and AWS offers no meaningful S3 commitment instrument; observed adders of 3 to 8 points suggest otherwise for estates where S3 is a large share of net spend. The test is dollar impact: below roughly 15 percent of net bill, spend the negotiating capital on egress or compute instead.