Our Three Year Spend Commitment Quietly Became an Architecture Decision

Two years ago the finance team and the platform team negotiated a discount with our cloud provider. In exchange for committing to a minimum annual spend for three years, we got a meaningful percentage off the list price. It was a good deal on the terms it was judged by, and I signed off on the technical side of it without much thought.

The consequence showed up later, and not on a bill.

When a team proposed moving a batch workload to a different provider that was genuinely better suited to it, the conversation was not about suitability. It was about the shortfall. Moving that workload would take us below the committed spend, and the penalty for missing the commitment cancelled out the saving. When another team wanted to run a component on hardware we already owned, the same arithmetic applied. The cheapest option for the organisation had become the option that kept the meter running, and every architectural review now had a finance clause in it that nobody had written down as a constraint.

The subtler effect was on optimisation. We had spent a year telling engineers to right size their instances, delete idle volumes and turn off development environments overnight. That work was still correct and we still wanted it, but underneath it there was now a floor: if efficiency reduced consumption too far, we paid for the gap anyway. The incentive to be frugal had been softened by contract, and people noticed.

I am not arguing against commitments. The discount was real, the workloads were stable, and a fair part of that spend was never going anywhere. What I would do differently is treat the commitment as an architectural constraint from the day it is signed rather than as a procurement outcome. That means sizing it against the portion of the estate we are confident will not move, not against the current total. It means recording it on the risk and constraint register that architects actually read. And it means that the renewal date is a design milestone, planned for a year ahead, rather than a renewal notice that arrives with six weeks left.

Contracts shape systems. Ours shaped a system for three years, and only one of the people who signed it knew that.

– Serguey Shinder