Every other chapter in this book rests on something published. This one cannot, and the honest thing is to say so in the first paragraph rather than the last.
No cloud or model vendor publishes the discounts it actually concedes. Negotiated rates sit under mutual NDA. There is no dataset of outcomes, no median concession by spend band, no public benchmark for what a given commitment is worth. Any chapter claiming otherwise is either recycling a consultancy's marketing or making it up, and a number invented here would be the most quotable thing on the page and the least defensible.
So this chapter contains no negotiated figures at all. What it contains is the part that is genuinely knowable: how to convert the public record — which this book has spent thirty-odd chapters assembling — into leverage. That is judgment reasoned from mechanism, and it is labelled as such throughout.
Your floor is already published
The single most useful fact about a negotiation you have not started: you already know the worst acceptable outcome, because the vendor published it.
From commitment maths, AWS's discount ladder for p5.48xlarge runs from 26.5% up to 62.4% at three-year All Upfront — publicly, self-service, available to anyone with a credit card. That figure is not a target. It is a floor, and it reframes the entire conversation:
Any negotiated offer that does not beat the published self-service rate for the same commitment is not a concession. It is the list price with a meeting attached.
This sounds obvious and is routinely missed, because negotiations are conducted in percentages off an unstated baseline. Insist that every offer be expressed as an effective hourly rate for a named instance in a named region, then compare it against the public ladder yourself. The vendor's own price list is the referee, and it is free.
The same discipline applies to model APIs. Published per-token rates, batch discounts and cache pricing are the baseline; an enterprise agreement has to beat the arithmetic in the output premium on your actual traffic mix, not on a generic blended rate.
Leverage is switching cost, and yours is measurable
The reason most cost negotiations go badly is that only one side has done the arithmetic. The vendor knows what you spend. It does not know what leaving would cost you — and neither, usually, do you.
Lock-in is a cost is the chapter that makes this concrete: egress, rewrite effort, retraining, and the operational risk of a migration are all estimable in advance. That estimate is your actual bargaining position, and it works in both directions:
- A low switching cost is leverage you can use. Not as a threat — as a fact you can state calmly, ideally with a competing quote attached.
- A high switching cost is leverage you should know you have already given away. If migrating would take two quarters, the vendor is negotiating with someone who cannot leave, and you should price your next architectural decision accordingly.
The practical consequence: the work that improves your negotiating position is engineering work, done months earlier. Portability, a second provider running something real, standard interfaces at the boundaries. By the time renewal arrives, your position is already fixed — the meeting only reveals it.
Read the counterparty's own filings
This is the one place where hard, checkable evidence is available, and it is routinely ignored.
Several of the vendors in the neocloud landscape are publicly listed and therefore file. Those filings disclose customer concentration, contract structure, remaining performance obligations and capex commitments — audited, under legal liability, and free to read.
That tells you things a sales conversation never will:
- How concentrated their revenue is. A vendor with severe customer concentration needs each large customer far more than a diversified one does. That asymmetry is disclosed.
- What their contracted backlog looks like. Remaining performance obligations describe how much revenue is already locked in and over what period — which speaks to how hungry they are for new commitments.
- What they have committed to spend. Capex and purchase obligations indicate capacity coming online that must be filled.
None of that is a negotiating figure. All of it is context that tells you whether you are the marginal customer or a rounding error, and those two situations call for entirely different asks. Read the filing before the meeting. It costs an hour and it is the only genuinely asymmetric information available to you.
What is structurally negotiable
Reasoning from how these agreements are constructed rather than from any disclosed outcome, the terms worth attention are not all the same kind of thing.
| Term | Why it is worth asking for |
|---|---|
| Rate for a stated commitment | The obvious one, and the one anchored by the public ladder |
| Commitment flexibility | Whether unused commitment rolls forward, or how instance families may be exchanged — this is where a forecast miss either costs you or does not |
| Term length | Shorter terms at similar rates are worth real money given how fast unit costs move |
| Egress and transfer | Directly reduces the switching cost above, so it compounds into future leverage |
| Capacity assurance | Often more valuable than price when supply is the binding constraint |
| Support and engineering access | Costs the vendor less than rate concessions, so it is frequently the easier yes |
The pattern worth noticing: several of these are worth more than the headline rate and cost the vendor less to grant. A team that spends all its negotiating capital on percentage-off and accepts a rigid three-year lock has usually optimised the wrong variable — particularly given that the break-even threshold in capacity planning depends entirely on flexibility surviving contact with a demand miss.
Things this chapter will not tell you
Stated plainly, because their absence is the point:
- What discount you should expect. Unknowable from public sources. Depends on spend, term, timing, region, and the vendor's quarter.
- At what spend level negotiation becomes available. Vendors do not publish thresholds and they differ.
- Whether a competing quote will move a given vendor. No public data exists.
- What anyone else got. Under NDA, and second-hand claims are unverifiable.
If you have seen these numbers asserted confidently elsewhere, the useful question is where they came from. In this book's experience of trying to source them, the answer is nowhere checkable.
The honest limit of this chapter
This entire chapter is reasoned from mechanism rather than measured — it is the only one in the book of which that is true, and it says so at the top rather than burying it here. The published discount ladder, the switching-cost components, and the existence of the disclosures are all sourced in the chapters cited. The claim that they constitute leverage, the classification of negotiable terms, and the advice about which to prioritise are judgment.
What survives the disclaimer is worth carrying anyway: you know your floor because it is published, you can measure your own switching cost, and your counterparty's financial position may be a matter of public record. Three facts, all checkable, none requiring anyone to tell you a number they are contractually forbidden to share.