One costed workload moved 10 to 20 percent; the whole estate threatened moved nothing

AWS discounts against the probability that something actually moves, not against the volume of the threat. That is why a named workload with a costed landing zone outperforms a strategy deck, and why the timing of the conversation matters more than its tone.

Prepared by Redress Compliance · August 15, 2026 · AWS advisory. Based on 25 to 35 AWS commitment and renewal negotiations, 2024 to 2025.

• Data egress at published rates, which is the line most plans either omit or estimate rather than calculate, per the egress brief.
• Re platforming and testing effort, priced with real engineering estimates rather than a percentage of the compute saving.
• A dual run period, during which both environments bill simultaneously and nobody is saving anything.
• Retraining and tooling change, which is where broad multi cloud programs quietly overspend by paying twice for the same capability.
• The shortfall exposure if leaving mid term triggers a commitment penalty, which is a cost of timing rather than of migration.
• Marketplace routing as a separate lever, since eligible purchases can count toward the commitment inside the 25 percent contribution cap.

The common advice is to threaten a full multi cloud migration and let the vendor react. In the negotiations we ran, broad threats with no costed plan moved price by almost nothing, because the account team could see the bluff. What moved 10 to 20 percent was narrower: one or two workloads genuinely portable, the exit costed in full including egress, and a single credible option anchoring the discussion.

The part buyers underestimate is timing rather than substance. Commitments are designed to reward concentration, which means every discount you accept converts future optionality into present savings. That is a legitimate trade and it is not reversible. Once the commitment is signed, your leverage is limited to whatever flexibility terms you negotiated up front, and the portability work you do in year two produces nothing except a better position at the next renewal. The window is open before each commitment and at renewal, and it is closed everywhere in between.

This also explains why costing the exit honestly serves you rather than the vendor. A plan that cannot survive its own arithmetic is worse than no plan, because presenting one and having it dismantled costs credibility you will need later. In 30 to 50 percent of the naive cases we reviewed, transfer and re platforming costs erased the apparent savings entirely. Note the distinction from the exit cost question: leaving is usually cheap measured against a year of committed spend, and a migration is often expensive measured against the compute saving that motivated it. Both are true, and only the second one decides whether your alternative is real. The exit economics sit in the lock in brief.

What holds price over time is therefore an operating choice, not a negotiating one. Keep one or two reference workloads genuinely portable, refresh the exit cost model each year, and give cloud economics a named owner. Without that owner, portability decays through ordinary engineering decisions that nobody made in bad faith, and the leverage disappears before anyone notices it has gone. Then time the conversation before the next commitment, where it is worth something. The commitment sizing sits in the renewal strategy, and the year round controls in the vendor management playbook.

Across roughly 25 to 35 AWS commitment and renewal negotiations in 2024 and 2025, credible alternatives moved price and empty threats did not:

The patterns: threats without arithmetic, portability assumed rather than maintained, and leverage saved for a moment inside the term when it no longer existed.

The buyer side move is one real option, costed and timed. The wider library sits in the AWS practice.

Does multi cloud actually create AWS negotiation leverage?

Only when a specific workload can move at a known cost. AWS prices against real risk, so a vague threat does not change a discount curve. Buyers with one costed exit workload secured 10 to 20 percent more at renewal than buyers with none, and the alternative does not need to be your whole estate.

What makes an alternative credible?

Three traits: a named workload that is genuinely portable, a costed landing zone on a second cloud, and a realistic timeline with a named owner. Remove any one of the three and an account team reads the position accurately as a bluff, then holds its price.

When is the leverage window open?

Before each commitment and at renewal. Commitments reward concentration with discount, so they shrink your future leverage the moment they are signed. Inside a term your leverage is limited to the flexibility terms you negotiated up front, which is why those terms matter more than the headline rate.

Why does egress decide whether an exit is credible?

Because it changes the comparison. A migration that looks cheaper on compute can cost more once data transfer, re platforming, dual running and retraining are counted, and 30 to 50 percent of naive plans were erased that way. Model the full exit before presenting it as an alternative.

Does a partial exit still help?

Yes, and it is the recommended posture. Moving even one workload proves portability and changes the negotiation without the cost of a full migration. The point is demonstrated capability rather than scale, since one workload that has actually moved settles the question a plan cannot.

How does Marketplace routing fit the strategy?

Eligible Marketplace purchases can count toward the commitment, which turns third party software spend into commitment fuel and can reach a discount tier sooner. The contribution toward commit is capped at 25 percent, and the risk is overcommitting to reach a tier you did not need.

What posture holds price over time?

Keeping one or two workloads genuinely portable at all times, with the exit cost model refreshed yearly and a named owner for cloud economics. It is an operating choice rather than a negotiation stunt, because portability decays quietly and takes the leverage with it.

Is broad multi cloud worth the overhead?

Rarely. Blanket multi cloud makes teams pay twice for skills and tooling, and it usually costs more than the concentration it was meant to cure. Surgical portability on one or two workloads delivers the negotiating benefit at a small fraction of the operating cost.