Cloud infrastructure is priced in two different languages, and most buying mistakes in this category come from comparing across them as if they were one. Hyperscalers publish meters: per service rates that look precise and predict almost nothing about your bill. Provider clouds quote subscriptions: flat numbers that look opaque and behave predictably for years. The same workload can be priced both ways, and the difference between those two numbers is usually the real decision. This guide explains how each language works, so the quotes make sense when they arrive.
The two pricing languages: metered and flat
A hyperscaler bill is the sum of hundreds of small meters: compute by the hour or second, storage by the gigabyte month, requests, transfers, and features each ticking separately. The published rates are real, but the bill is the product of rates you can read and usage you have to predict, and the second half is where predictions fail. A provider cloud inverts the deal: a quoted flat rate per workload, per virtual machine, or per resource pool, steady every month for the term. You trade the meter's elasticity for a number that behaves. Neither language is wrong; they reward different workload shapes, which is why the comparison has to start with your workloads rather than the price lists.
What the meters actually charge
Three meters dominate most bills. Compute is the visible one, and in 2026 it moved in two directions at once: ordinary processing got cheaper while anything touching AI capacity got scarcer and pricier, so the workload mix now decides how the economics feel. Storage is the quiet accumulator, growing a little every month and rarely reviewed. Egress is the one that surprises: moving data out, whether to your users, another provider, or your own backup, carries a toll that only shows up once real volumes flow. Egress deserves particular attention because it prices the exit. A workload that is cheap to run and expensive to leave is not cheap; it is committed.
Commitments are where the money moves
Nobody with a real estate pays the on demand rate for everything. Hyperscalers discount meaningfully for committed use, by term and by prepayment, and larger estates negotiate enterprise agreements on top. The catch is symmetry: a commitment sized from measured usage is the easiest money in the category, and a commitment sized from optimism becomes a bill for capacity that never ran. Provider clouds carry the same logic in contract form, with one to three year terms and the price protections written at signing. In both languages the discipline is identical: commit to the floor you can prove, keep the spiky remainder flexible, and put the renewal date on a calendar.
The VMware licensing line
For VMware estates, licensing has become its own line on every cloud quote. The upheaval concentrated licensing into bigger bundles at higher committed spend, which lands hardest on companies running modest estates on their own hardware. Provider clouds that hold top standing in the VMware program carry the licensing at platform scale and fold it into the quoted rate, which is a large part of why their numbers have become newly competitive against both self hosting and rebuilding elsewhere. When comparing those quotes, confirm what the licensing line covers at your host count today and what happens to it if the program's terms move again; the strongest providers will answer in writing.
The managed layer prices on top
Management is priced separately from infrastructure in either language, whether it appears as a line item or is folded into the rate. What it buys is the operating work: monitoring, patching, cost governance, incident response, and the engineering judgment that keeps a metered bill from drifting. The honest comparison is against your alternative cost, which is hiring and retaining the same skills in a market that prices them aggressively. Thin teams with steady estates usually come out ahead buying the outcome; deep teams with real platform investment usually keep it in house. What rarely works is the middle: a metered platform, no cost discipline, and nobody owning the bill.
Comparing total cost honestly
Put every quote in the same shape before comparing: your actual workload profile rather than a generic one, egress modeled at real volumes, backup and disaster recovery included, licensing stated, the management scope declared, and commitments matched to measured usage. Then read the three year total, not the monthly line. On that math the published meter frequently loses to the quoted flat rate for steady estates, and wins for genuinely elastic ones, which is the honest answer to most pricing questions in this category. Our ranked evaluation of the cloud infrastructure market states how each provider's pricing behaves, and our guide to buying cloud infrastructure covers the decision sequence around the number. Reshaping quotes into that comparison is exactly the work we do for buyers across the whole market, free, and you sign directly with whichever provider you choose.
Frequently asked questions
- How much does cloud infrastructure cost per month?
- Honest answer: at any real size it is workload dependent, and any single figure quoted without a workload profile is a teaser. Hyperscaler bills are the product of published rates and your actual usage, moved substantially by committed use discounts. Provider clouds quote flat monthly rates per workload or resource pool for a term. The meaningful number is the three year total for your specific estate, with egress, backup, recovery, licensing, and management included, modeled on both paths. That modeling is the work we do for buyers, free.
- Why is our cloud bill so unpredictable?
- Because a metered bill is a prediction, and the meters most likely to run past it are the ones nobody watches: storage that accumulates, egress that scales with success, and services someone turned on for a project that ended. The fixes are unglamorous: commitments sized to the measured floor, alerts on the meters that move, someone who owns the bill by name, and a periodic review that turns things off. If that discipline is not realistic for your team, that is not a character flaw; it is a signal that a flat rate provider cloud or a managed layer fits your operation better than the meter does.
- Is a provider cloud cheaper than a hyperscaler?
- For steady state workloads, frequently yes, once the full bill is compared: flat quoted rates against metered compute, storage, egress, and the licensing line, plus the management the meter assumes you will do yourself. For genuinely elastic workloads, usually no, because you would be paying a flat rate for capacity you only need in bursts. Most real estates contain both kinds, which is why the honest answer is usually a split rather than a winner, and why the comparison has to be run on your actual workload profile rather than a benchmark.
- What should we commit to upfront?
- The floor you can prove, and no more. Measure a real usage baseline, commit to the steady portion that clearly is not going away, and leave the variable remainder on flexible terms even though it prices higher per unit. Oversized commitments are the most common self inflicted wound in this category, because the discount on paper becomes a bill for capacity that never ran. The same rule reads across both languages: hyperscaler commitments and provider cloud terms both reward accuracy over ambition, and both deserve a renewal date on your calendar.