The commit structure and the egress line set the bill, not the rate card
Cloud cost is set by the commitment structure and the data transfer line, not by the published rate card: the realized discount on a well sized three year commit lands in the 8 to 14 percent band rather than the 25 percent headline tier, because unutilized commit, egress, and non discounted services drag the blend back. AWS, Azure, and Google Cloud price commits differently and converge on the same realized band once usage is normalized, which means the realized cost is a function of how the buyer structures the deal, not which logo is on it.
Prepared by Redress Compliance · August 8, 2026 · Cloud advisory. Based on 60 to 80 hyperscaler renewals and commit sizing engagements supported 2024 to 2025.
Executive summary
The vendor proposed commit ran 22 to 38 percent above the trailing draw.
The opening commitment proposal was rarely the right one: the vendor proposed commit ran 22 to 38 percent above trailing twelve month draw on the median estate, before any growth assumption.
And over committing to chase a deeper tier is the costliest common mistake, because the effective rate ends higher, not lower, once the unutilized portion pays the headline price for nothing.
The sizing rule is trailing twelve month draw plus a defensible growth band, never the forecast.
Egress took 12 to 22 percent of bills that budgeted under 5. Internet egress and inter region transfer accounted for 12 to 22 percent of the realized bill, against an average buyer budget of under 5 percent: internet egress at 5 to 9 cents per gigabyte prices small per unit and large at scale.
Inter region and cross zone transfer compound silently in multi region designs, and the line appears in no commitment vehicle, which is why it drives more overrun than any other unit cost.
Egress budgets as a first class line, and architecture, not negotiation, controls it.
The headline tier and the realized band are different numbers everywhere.
The commit ladders run AWS EDP five year deep tiers above 25 percent, three year mid tiers at 12 to 18, Azure MACC at 10 to 16, and GCP committed use at 10 to 17.
And the realized discount after the unutilized commit is counted lands at 8 to 14 percent effective across all of them: the mechanics differ, the leverage points differ, and the trap converges, with a poorly structured Azure commit out costing a well structured AWS commit on identical workloads.
The realized cost index ran 100 in 2021 to 146 in 2026, steepening as priced AI services arrived.
AI consumption broke every forecast, and part of the bill sits at list no matter what. AI service consumption grew 3 to 5 times faster than the buyer forecast on every estate measured, pulling realized rates up regardless of commit tier.
And a structural share of the bill sits outside the commit entirely, on demand above the ceiling, managed service surcharges, and internet egress billed at list and rarely discounted at typical volumes: underestimating that uncontrolled share is the single most common budgeting error.
And it is controllable through architecture and traffic shaping, never through the discount sheet.
The commit ladder, headline against realized
| Commit vehicle | Headline list discount | The reality |
|---|---|---|
| AWS EDP, five year deep tier | 25 percent plus | The tier over committing chases |
| AWS EDP, three year mid tier | 12 to 18 percent | The workable band for most estates |
| Azure MACC, three year | 10 to 16 percent | Same shape, different clothes |
| GCP committed use, three year | 10 to 17 percent | Converging on the same band |
| One year plans and reservations | 4 to 10 percent | The flexibility premium priced |
| Realized, after unused commit | 8 to 14 percent effective | Where every well sized deal actually lands |
The gap between on demand and a three year commit can exceed 50 percent on compute, and few estates capture it.
Unutilized commit, egress, and non discounted services drag the blended figure back toward a 10 percent realized rate, and that gap is the entire game: the buyer who builds the negotiation around realized cost structures the deal.
And the buyer who chases the headline tier pays the headline price on the unutilized portion.
Two estates of identical spend can land 12 points apart purely on sizing and transfer mix.
The egress trap, engineered or paid
- Internet egress at 5 to 9 cents per gigabyte: small per unit, large at terabyte scale, and the line buyers forget because no commitment vehicle carries it.
- Inter region and cross zone transfer: the silent compounding cost of multi region designs, paid on every call the architecture routes across a boundary.
- The budget gap: 12 to 22 percent of realized bills against under 5 percent budgeted, the widest planning miss in the file.
- The control is architectural: traffic shaping, region consolidation, and data locality decisions, priced at design time in dollars per million calls.
- The negotiation reality: egress sits at list at typical enterprise volumes, so the discount sheet never fixes what the architecture created.
The AWS EDP negotiation brief
The commit sizing method, the discount ladder mechanics, and the terms that survive a usage miss, worked across the vehicles.
Get the white paper →Sizing the commit, and where the clouds converge
The sizing discipline transfers across all three clouds because the trap does: commit to the trailing twelve month draw plus a defensible growth band, hold the AI consumption forecast separate and reviewed quarterly since it broke every estimate by 3 to 5 times.
And price the walkaway between vehicles, because AWS Savings Plans plus EDP, Azure reservations plus MACC, and GCP CUDs converge on the same 8 to 14 percent realized band once usage is normalized, making the structure, not the vendor, the variable.
The AWS specific recovery program runs in the AWS overspending report, the egress negotiation positions in the egress negotiation guide, the OCI variant of the same commitment arithmetic in the OCI cost analysis, and the AI meters pulling every forecast upward in the token cost surge report.
- 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
What we saw across cloud engagements, 2024 to 2025
Across roughly 60 to 80 hyperscaler renewals and commit sizing engagements our team supported between 2024 and 2025, three patterns recurred on enterprise estates:
Vendor proposed commits above trailing twelve month draw, before any growth assumption.
The blended realized cloud cost path from 2021 to 2026, steepening as AI services priced in.
The benchmark's one transferable sentence covers all three logos: realized cost is a function of how the buyer structures the deal, not which cloud runs it, and the three levers, the commit sized to measured draw, the egress engineered as a first class line.
And the AI forecast held separate and reviewed quarterly, are identical across AWS, Azure, and GCP because the traps are.
The uncontrolled share, on demand above the ceiling, surcharges, and internet egress at list, is the budgeting error that repeats most, and it yields to architecture and traffic shaping while remaining forever immune to the discount conversation.
Your first five moves
- Size the commit to trailing twelve month draw plus a defensible growth band, against proposals running 22 to 38 percent high.
- Budget egress as a first class line, the 12 to 22 percent that sub 5 percent budgets kept missing.
- Refuse the deeper tier chase, since the realized band is 8 to 14 percent regardless of the headline.
- Hold the AI forecast separate and review it quarterly, because it grew 3 to 5 times faster everywhere.
- Engineer the uncontrolled share at design time, the list priced remainder no discount reaches. The cost optimization practice runs the sizing with you.
Frequently asked questions
What discount do enterprises really get on cloud commits?
The realized band on a well sized three year commit runs 8 to 14 percent below blended list, not the 25 percent plus headline tier: unutilized commit, egress, and non discounted services drag the blend back.
And the gap between on demand and committed rates that can exceed 50 percent on compute is captured by few estates.
AWS, Azure, and GCP converge on the same realized band once usage is normalized.
How badly are cloud commits oversized?
The vendor proposed commit ran 22 to 38 percent above trailing twelve month draw on the median estate, before any growth assumption, and over committing to chase a deeper tier is the costliest common mistake: the hours below the commit line pay the committed rate for nothing.
And the effective rate on what actually ran ends higher than a smaller commit would have produced.
How much does cloud egress really cost?
12 to 22 percent of the realized bill in our engagements, against an average buyer budget of under 5 percent: internet egress at 5 to 9 cents per gigabyte scales brutally at terabyte volumes, and inter region and cross zone transfer compound silently in multi region designs.
The line sits in no commitment vehicle and at list at typical volumes, so the control is architectural, priced in dollars per million calls at design time.
Which hyperscaler is cheapest on committed spend?
None, once usage is normalized: AWS prices through Savings Plans and the EDP layer, Azure through reservations and the MACC, and GCP through CUDs, and all three converge on the same 8 to 14 percent realized band.
The realized cost is a function of how the buyer structures the deal, and a poorly structured commit on any cloud out costs a well structured one on any other, on identical workloads.
How should AI consumption be handled in cloud commits?
Separately and skeptically: AI service consumption grew 3 to 5 times faster than the buyer forecast on every estate we measured, pulling realized rates up regardless of tier, so the AI line holds outside the base commit with its own quarterly review.
Folding an optimistic AI forecast into the commitment is how the 22 to 38 percent oversizing happens twice.
What part of the cloud bill cannot be discounted?
The structural remainder: on demand usage above the commit ceiling, certain managed service surcharges, and data transfer to the internet, all billed at list and rarely discounted at typical enterprise volumes.
That share is controllable through architecture and traffic shaping but immune to negotiation, and underestimating it is the single most common cloud budgeting error in the file.
