The vendor’s solutions architect knew our system better than two of my own engineers did. He had read our reference architecture, the one we sent over for “sizing,” and he quoted it back to us in the meeting to explain why a particular managed service was the obvious choice. He was not wrong. That was the problem.
This is what a nine-figure cloud commitment actually is. Not a discount you sign for. A bet that your spend will keep climbing, sold to you by people who have studied exactly how to make sure it does. I have sat on the buy side of one of these, a multi-year commit across compute, storage, data egress, and a long tail of managed services, on a platform doing billions in volume. I went in thinking it was a finance exercise. It is an engineering one, and the engineers were almost not in the room.
What you are actually buying
The headline is a number and a percentage off list. Commit this much spend over this many years, get this much discount. Simple. The simplicity is the trap.
What you sign is a floor, not a ceiling. You are promising to spend at least that much. If your architecture gets more efficient, if a team kills a wasteful pipeline, if you migrate a workload to something cheaper, you still owe the floor. I have watched a cost-optimization win get quietly shelved because hitting the savings would have put us under our commit, and the penalty for missing the commit was worse than the waste. Sit with that. We were structurally disincentivized from being efficient. (Nobody draws that on the architecture diagram.)
The tiers make it worse. The deeper discount sits one bracket up, always. You committed to a number that felt like a stretch, and within a quarter the account team is showing you how close you are to the next tier and what it would buy. The pressure to keep spending to hit a tier is not a side effect. It is the mechanism.
Where engineers have to be
Here is the part nobody tells you. The salespeople will quote your own architecture back at you, and if no engineer is in the room, that architecture becomes the contract.
A solutions architect’s job is to map your discount to the services that are hardest to leave. Your discount on raw compute is real and portable. Your discount on the proprietary managed database, the serverless runtime with the bespoke event model, the data warehouse with the query dialect that exists nowhere else, that discount is bait. It is cheaper today and it welds you to the platform for the length of the commit and well past it. Finance cannot see the difference. The line items look the same. An engineer who has tried to migrate off a managed service knows that one of them is a door and the other is a wall.
So the engineering question in a procurement meeting is not “what is the price.” It is “what does this price cost me in optionality.” Every deep discount on a sticky service is the vendor paying you, once, to never leave. Sometimes that trade is fine. You should at least know you are making it.
Portability is the only card you hold
A commitment is a negotiation, and you have exactly one source of power: the credible ability to walk. Not the whole platform. Individual workloads.
The discount you get is roughly proportional to how believable your alternative is. If everything you run is welded to one vendor’s proprietary surface, you have no alternative, and the account team can read that in your architecture as clearly as a credit score. If a meaningful share of your fleet runs on portable substrate, containers, open formats, standard data layers, Postgres-compatible engines, Iceberg tables instead of the proprietary warehouse, then your threat to move a workload is real, and real threats move prices.
This is the uncomfortable part for engineers who love the slick managed service. The most elegant, most platform-native architecture is often the worst negotiating position. The boring portable one, a little more ops work, a little less magic, is what gives your CFO a card to play. I am not saying refuse every managed service. I used plenty. I am saying know which ones are load-bearing for your independence and keep those portable on purpose, even when the native option is nicer. You are buying room to walk with a little inconvenience, and that room is the only thing on the table that compounds.
The discounts that matter and the ones that are theater
Most of what gets paraded in these deals is theater. Credits that expire before you can spend them. Training vouchers. Co-marketing. “Executive sponsorship,” which means a smart person from the vendor will sit in your meetings and learn your roadmap. Architecture review days, which are useful and are also reconnaissance. None of that touches your unit economics.
The discounts that matter are the ones on the things you provably cannot avoid spending. Steady-state compute. Egress, which is the toll booth nobody negotiates hard enough and the single biggest lever in a multi-cloud posture. Storage at the volume you will actually hold. Get those down and the deal is real. Everything else is a number designed to make the total look generous while the spend that counts stays close to list.
And read the ramp. A commit that starts gentle and balloons in years two and three is betting on your growth, and if the growth does not show, you are buying capacity you do not need at a discount that no longer feels like one. I have seen a team make next year’s roadmap decisions to defend a commitment signed two years earlier by people who had since left. The contract outlived the strategy. That is the failure mode: the deal stops serving the architecture and the architecture starts serving the deal.
The room
Get an engineer who has done a migration into the negotiation. Not for the pricing model. For the one question finance cannot ask: which of these line items is a door and which is a wall. The vendor already knows your architecture cold. The least you can do is bring someone who knows it better.