Negotiating AWS Data Transfer: Rate Card, Architecture, Exit Leverage

List internet egress still opens at $0.09 per GB in 2026, while the EU Data Act drives switching charges to zero on January 12, 2027. Your negotiation lives between those two numbers.

Prepared by Redress Compliance  ·  June 2026  ·  Representative AWS estate scenario (benchmark scenario, not a quote)

You cannot negotiate a line you cannot decompose. AWS data transfer is not one price. It is at least seven distinct meters, and they stack on the same traffic.

Figure 1. List cost per GB by AWS transfer path, 2026 rate card. CloudFront private rate is an engagement file benchmark, not a published price.

Three mechanics on this card are routinely missed in contracts we review. First, the NAT toll stacks: a gigabyte exiting through a NAT gateway pays $0.045 processing plus the DTO tier rate.

Second, cross AZ is billed on both sides, so the real rate is $0.02 per GB moved. Third, DTO tiers reset monthly, which keeps seasonal traffic re entering the expensive first tiers.

AWS controls the renewal calendar, the reference prices, and the consumption data presentation. The buyer side framework flips each of those in a twelve week sequence run ahead of any EDP signature, renewal, or private pricing request.

The order is the point. Buyers who negotiate first and optimize second hand AWS an inflated baseline, then spend the term underconsuming a commitment sized on traffic they since eliminated.

A verified baseline survives AWS scrutiny when it is built from the Cost and Usage Report at line item grain, not from Cost Explorer screenshots. CUR is the billing system of record. It lets you attribute every transfer dollar to a path, an account, and a workload owner.

Here is the benchmark estate we use to illustrate the method: a media and analytics business pushing 800 TB to the internet monthly (benchmark scenario, not a quote; the free 100 GB monthly tier is excluded as immaterial).

That is $936,000 a year, and on the first pass AWS will frame it as the cost of doing business. The baseline work proves otherwise. Roughly 80 percent of the NAT processing in this estate is S3 bound traffic that a free gateway endpoint eliminates outright.

CloudFront fronting moves all 800 TB of DTO off the $0.09 ladder. The origin fetch is free, and committed CDN volume prices at a $0.02 blended benchmark under private pricing. Topology fixes halve the cross AZ chatter, and replication trimming cuts inter region volume by a third.

Figure 2. Benchmark estate monthly transfer cost by category. NAT gateway hours ($200) unchanged and omitted from bars. Benchmark scenario, not a quote.

The standard reseller and account team advice is to sign a larger EDP and let the headline discount absorb the egress line. We disagree, and the engagement data is why. The EDP percentage discounts a rate card you can collapse architecturally first, and it creates a consumption obligation in the same stroke.

Here is the mechanic most buyers learn too late: data transfer spend draws down your EDP commitment. Cut egress 63 percent mid term, as the worked estate does, and the saved dollars reappear as commitment shortfall exposure at true up. Your own optimization becomes a penalty.

The architecture levers themselves are not exotic. Gateway endpoints for S3 and DynamoDB are free and remove the NAT toll. PrivateLink replaces NAT for service to service traffic at roughly a quarter of the processing rate. CloudFront fronting converts $0.09 tier DTO into CDN rates AWS will discount much more aggressively.

Discounts decay; clauses persist. These five decide whether your commitment protects the budget across the term. None of them appear in AWS first draft paper.

These are the ranges we see when the instruments above are negotiated against a verified baseline. They come from buyer side engagements, not vendor collateral. Benchmark ranges: Redress Compliance advisory engagement file, 2024 to 2025.

Figure 3. Discount ranges by scenario, matching the table above. Benchmark ranges: Redress Compliance advisory engagement file, 2024 to 2025.

Two reading notes. The DTPP and CloudFront ranges require committed monthly volume, typically with a 12 month floor, so they belong after the baseline work, not before. And the exit row is leverage even if you never use it: a credible, priced exit reframes every other row.

Egress fees were designed as exit friction, and that design is now failing. Since March 2024 AWS has provided free data transfer out for customers leaving the platform, through credits arranged via AWS support for accounts moving all data off. The EU Data Act goes further: switching charges must be cost based today and disappear on January 12, 2027.

Be precise about what the public program is not. It is not a steady state multi cloud subsidy. It requires a full migration, an application to AWS, and it arrives as credits, not a rate change. As an operating lever it is nearly useless. As negotiation leverage it is excellent, because it prices your exit at zero.

The BATNA file should price your top egress driving workloads on at least one alternative: zero egress object storage providers, another hyperscaler with free switching terms, or a colocation plus CDN design. The point is not to move. The point is that AWS believes you can.

Pair the side letter with clause 4 from Section 5 so the Data Act regime applies globally. AWS contracting entities outside the EU have no obligation to extend it, which is exactly why you put it in the paper now, while it still reads as a concession.