The AWS, Azure, and GCP Coordinated Competition Framework

Uncontested cloud renewals in our file landed at 7 to 12 percent off list. Coordinated competition landed at 18 to 26 percent. The EU Data Act abolishes switching charges by January 12, 2027, and the leverage window is open now.

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

AWS holds roughly twice the workload share of Azure in most enterprise estates we benchmark, and the account team negotiates like it. The default cycle has AWS controlling the calendar through commit expiry, the reference points through its own TCO decks, and the forecast through your engineers.

Yet 2024 to 2026 shifted leverage toward buyers. All three vendors waived exit egress fees, Google first in January 2024 and AWS and Microsoft within weeks. The EU Data Act then put switching rights into law, and exit friction, the historical anchor of hyperscaler pricing power, is being legislated away.

At the same time, GenAI spend gave the vendors a new commit inflation instrument. Every renewal we have seen since 2024 arrives with an AI consumption forecast attached. Section 5 covers how to neutralize it.

Coordinated competition is not a threat to leave. It is a sequenced program that makes three vendors bid for defined workload families on your calendar. The framework runs in three phases over roughly twelve weeks, and the order is the discipline.

The framework holds whichever cloud is your anchor. We run the same sequence for Azure heavy estates with AWS as the challenger; the tactics in section 11 simply change vendor names.

Cloud has no license entitlements in the classic sense. The asset the vendor scrutinizes is your forecast, so the baseline must be built from metered fact. Three properties make it defensible when the AWS deal desk pressure tests it.

• Service level granularity: twelve months of consumption by service, account, and region from Cost Explorer and the rival consoles, with reserved, savings plan, and spot coverage separated from on demand drift.
• Waste accounting: idle capacity, unattached storage, oversized instances, and expiring reservations priced at list and presented as recoverable spend. Growth is funded from documented waste before it is funded from new commit.
• Portability scoring: every workload family tagged anchor, challenger, or edge, with a named owner and a migration cost estimate. This is what converts the baseline from accounting into leverage.

The three commit vehicles look interchangeable and are not. The differences below are where coordinated buyers extract terms that single cloud buyers never see.

The tier curve is the key shape. Bands in our file run roughly 7 to 12 percent under $5M annual commit, 12 to 18 percent to $20M, 18 to 22 percent to $50M, and 22 to 28 percent above. The curve flattens hard at the top: the last $20M of concentrated spend often buys two points or less.

Ramp discipline: commit year one at your verified run rate, not the blended average of an optimistic three year forecast. A flat $30M commit against a $24M year one run rate is a $6M unsecured loan to your own growth plan.

AI spend rolls into all three commit vehicles: Bedrock consumption counts in full toward an AWS EDP, Azure OpenAI consumption draws down a MACC, and Vertex AI burns a Google commit. That rollup is genuinely useful, because AI spend inherits your negotiated discount.

It is also the inflation instrument. The account team's AI forecast justifies a bigger commit and a longer term, and the shortfall clause means you pay it whether the AI roadmap lands or not. Three rules keep the overlay safe.

• Commit to measured AI consumption only. Structure forecast upside as optional tranches that join the commit when consumption proves out, never as committed floor.
• Keep provisioned AI capacity out of the base commit. Bedrock model units, Azure OpenAI provisioned throughput units, and Vertex provisioned throughput are separate fixed obligations that bill used or not. Buy them against measured traffic floors on the shortest term.
• Use AI portability as challenger leverage. Inference behind a gateway is the most portable workload family in the estate. A five percent live traffic split to a second provider converts the AI forecast from vendor leverage into buyer leverage.

Egress fear is the oldest objection to coordinated competition, and it is mostly stale. Since early 2024, Google, AWS, and Microsoft all waive data transfer out for customers leaving the platform. Conditions apply: roughly 60 day transfer windows, approval through support, and on Azure, closure of the associated subscriptions.

Read the conditions carefully: the waivers fund a full exit, not an ongoing multi cloud architecture. Steady state transfer between clouds still bills at list, around $0.09 per GB on AWS internet egress tiers after the free allowance. Architect challenger workloads to minimize cross cloud chatter, and the objection collapses.

The marketplace mechanic is the asymmetry most buyers never price. Third party software bought through each cloud's marketplace can retire your commit, but the three vendors count it very differently.

The pull through play: an estate spending $8M a year on eligible third party software can burn a MACC dramatically faster than an EDP. Route that spend deliberately. In an AWS led negotiation, the rival marketplace terms belong in the rival term sheet, priced and presented, because they change the effective cost of every challenger scenario.

The shortfall clause is the entire value of the commit to the vendor. On all three clouds, an unconsumed commitment is invoiced at term end, and rollover of unspent commit into a renewal is a negotiated exception, never a default. Ask for carryforward in writing; AWS grants it most readily when a renewal is signed alongside.

Overage runs the opposite trap. Consumption above the commit usually bills at your discounted rate, which feels safe, so estates drift far above commit and arrive at renewal with no headroom story. The vendor reads sustained overage as guaranteed growth and prices the next term accordingly.

The renewal posture that works: arrive 10 to 20 percent over commit with the overage attributed to named, portable workloads. You demonstrate growth, and you demonstrate that the growth has somewhere else to go.

Six failures recur across the estates we are brought into after the fact.

• Committing to the forecast, not the run rate. The shortfall clause converts optimism into invoice.
• Letting the incumbent see a finished migration plan too late. Leverage peaks while the challenger award is still open.
• Splitting workloads evenly across clouds. See the contrarian note below; even splits buy worst tier pricing twice.
• Ignoring the marketplace asymmetry. Routing eligible third party spend through the wrong cloud strands commit burn worth points of effective discount.
• Signing AI capacity into the base commit. Fixed throughput obligations survive the workloads they were bought for.
• Treating professional services credits as discount. Credits expire, carry conditions, and anchor you to the vendor's delivery teams. Price the deal as if they did not exist.

Where the common advice on multi cloud splits is wrong. The standard consulting pitch is to split workloads broadly across two or three clouds for resilience and leverage. We disagree: estates that split near 50/50 in our file landed in lower discount tiers on both vendors and paid more in aggregate than concentrated estates.

The structure that wins is roughly 70 percent anchor, 20 percent challenger, 10 percent edge: the anchor keeps its tier position, and the challenger segment is large enough that both vendors believe it can move. Leverage comes from credible mobility at the margin, not from symmetry.

Discount percentage is what the vendor wants you to negotiate. These five clauses are what decide whether the number survives contact with three years of reality. Every one has been accepted, in some form, in deals we have advised.

Priority order matters. Clauses 1 and 2 carry the money. Clause 5 is usually the cheapest to win because it codifies what the vendor already claims publicly, and it is the clause that keeps your next negotiation honest.

Discounts track demonstrated mobility, not spend alone. The bands below are drawn from engagements across our advisory file, normalized as effective discount against list for the whole estate.

The worked estate makes the spread concrete. Take a representative $30M annual cloud spend, $90M over a three year term, AWS as anchor. The vendor opening lands at 12 percent, a well run single cloud negotiation reaches 18 percent, and the coordinated posture reaches 26 percent.

Benchmark ranges: Redress Compliance advisory engagement file, 2024 to 2025.

AWS account teams run a consistent sequence around competitive renewals. None of it is improper; all of it is directional. The counters neutralize each move without burning the relationship, and the same plays appear in Microsoft and Google colors.

A BATNA in this negotiation is not a slide that says "Azure exists." It is a priced, dated, partially executed alternative. The build has three parts, and the first two are cheap.

• Two written rival term sheets covering the challenger segment, with migration funding, marketplace terms, and committed discounts, refreshed inside 90 days of the anchor close.
• One live second path: a production workload family, however small, running on the challenger cloud with real spend. Operational proof converts the term sheets from hypothetical to priced reality.
• A signed internal decision rule: the board approved threshold at which the challenger award grows. When AWS knows a number exists, even without knowing the number, behavior changes.

Then write the protections into a side letter. Sample language we use as the starting position:

You will not get every sentence. Each one you do get converts a program page or a sales assurance into contract. The rate card is public, the protections are not, and only one of them is negotiable.