The commitment had been signed on the hyperscaler's growth forecast, not the customer's. Two years in, actual consumption trailed the commit by a wide margin — and the customer was paying for the gap.
The opening from the vendor was that commitments are fixed. They are not. Migration credits, a re-based ramp, and a re-sized commit are all negotiable when the alternative — a credible second hyperscaler — is on the table.
We modelled the real trajectory, sized a defensible commit to it, and traded a longer term for a lower unit price and drawdown flexibility. The forecast gap stopped being the customer's problem.
The lesson: a cloud commitment is only as good as the forecast under it, and the forecast is always negotiable later if you keep optionality alive.