Cloud computing was sold on a simple proposition: replace capital expenditure with a metered utility bill and pay only for what you use. The proposition is real, and for variable workloads it is transformative. But the contracts that govern large enterprise cloud relationships have grown into instruments with pricing structures closer to airline revenue management than to a utility, and the gap between headline rates and realised cost is where most cloud budget overruns live.
The commitment discount trap
Every major provider offers substantial discounts in exchange for committed spend over one to three years, and finance departments like them because they convert an unpredictable bill into a planned one. The complication is that commitments are usually denominated in dollars rather than in specific resources, and they do not shrink if the workload does. An organisation that commits based on an optimistic growth forecast and then consolidates, migrates or simply becomes more efficient can find itself paying for capacity it no longer needs while being told the commitment is the reason its rate is favourable.
Read moreHow Drug Prices Start in the Research Lab
Reserved instance and savings-plan structures add a second layer, because they are specific to instance families and regions. Buying three-year reservations for a compute generation that is superseded eighteen months later locks an organisation onto older hardware at a rate that new customers beat on demand.
Egress and the gravity of data
Moving data into a cloud is generally free. Moving it out is generally not, and at scale the charge for data transfer out can exceed the cost of the storage itself. The practical effect is a switching cost that grows with every terabyte accumulated, which competition authorities have begun to treat as a market-structure question rather than a pricing detail. Regulatory reviews of cloud market competition, including work by the Federal Trade Commission, have examined exactly this dynamic, and several providers have since waived transfer fees for customers leaving entirely.
Egress is only the visible part of data gravity. Managed database services, proprietary serverless runtimes and identity integrations all raise the engineering cost of migration far above the transfer bill, and those costs never appear on an invoice.
Read moreThe Real Economics of Corporate Layoffs
Licensing inside the bill
A significant share of enterprise cloud spend is software licensing rather than infrastructure, and licence terms for commercial databases and operating systems frequently differ depending on which cloud they run in. Buyers who negotiate infrastructure rates hard and accept bundled licensing terms unexamined often find the second number is larger. Public-sector buyers face the same problem with more transparency; procurement guidance and audit findings from the Government Accountability Office document repeated cases of agencies unable to reconcile cloud spending to delivered capacity.
What good governance looks like
Organisations that control cloud cost successfully tend to do four unglamorous things. They attribute every dollar to a team and a product, so that cost is visible to the engineers who create it. They review commitments quarterly against actual consumption rather than annually against the original forecast. They require an exit assessment for any managed service adopted, documenting what migration away would cost. And they treat architecture decisions as financial decisions, because the choice between a managed service and a portable one is a trade of engineering effort against future negotiating leverage.
None of this argues against cloud adoption. Standards work on interoperability and portability, including guidance published by the National Institute of Standards and Technology, exists precisely because the model works well enough to be worth making competitive. It argues instead for treating the contract as a strategic document rather than a procurement formality.
Read moreHow Fed Rate Decisions Move Your Portfolio
The clauses that determine the real bill
The headline rate in a cloud agreement is rarely where the money is. The provisions that matter are the ones governing data egress, committed-use discounts, support tiers, and the definition of the unit being billed. Egress charges mean that moving data out costs materially more than moving it in, which converts an apparently portable workload into a sticky one. Commitment discounts trade price against flexibility over one to three years, and they are priced on the assumption that most customers overestimate their ability to forecast consumption. Support that is adequate for production is usually a percentage of spend rather than a fixed fee, so it scales with the very growth the contract is meant to enable.
Guidance on managing this is now mainstream practice rather than specialist knowledge; the FinOps Foundation publishes frameworks for attributing cloud cost to teams and for negotiating commitments against measured rather than assumed usage. The organisations that control cloud spend well share one habit: they instrument consumption before they negotiate, so the commitment conversation is grounded in data the vendor cannot dispute.
