Updated 6 Oct 2026 • 5 mins read
CapEx vs OpEx in the Cloud: Models, Formulas, and Examples for SaaS Teams
CapEx is money spent up front on owned assets and depreciated over years; OpEx is ongoing spend expensed as incurred. Cloud shifts infrastructure from the first to the second. This guide explains both models, gives the formulas to compare them fairly, works through a US example, and covers why cloud commitments blur the line.
Twenty years ago, buying infrastructure meant buying it: servers arrived on pallets, sat on the balance sheet as assets, and lost value on a schedule the accountants chose. Cloud changed that into a monthly bill for capacity you never own. Finance teams call the first model CapEx and the second OpEx, and the shift between them is one of the most consequential and least understood effects of moving to the cloud, because it changes not just how much you spend but when it hits the books, what it does to margins, and who in the company gets to decide.
This guide explains both models for the people who actually have to compare them: engineering leaders choosing where workloads run and SaaS finance teams explaining the result to a board. It covers what each term means in accounting and in practice, gives the formulas for a fair comparison including the utilization adjustment most comparisons omit, works through a US example that flips depending on one variable, and covers the nuance that makes cloud spend less purely OpEx than it looks. We are not accountants and this is not accounting advice; the specific treatments below should be confirmed with yours.
The short answer CapEx (capital expenditure) is spending on assets you own, such as servers, recorded on the balance sheet and depreciated over several years, typically three to five for IT hardware. OpEx (operating expenditure) is recurring spending on services you consume, such as cloud compute, expensed in the period it occurs. Cloud computing shifts infrastructure from CapEx to OpEx: no upfront purchase, no depreciation, a monthly bill that scales with usage. The trade is cash preservation and flexibility in exchange for a recurring cost that sits in cost of revenue and directly affects SaaS gross margin. Cloud is cheaper for variable or uncertain demand because on-premises capacity bought for peak sits idle; on-premises can be cheaper for steady, high-utilization workloads over a full hardware life. Compare them on total cost of ownership per unit of output over the same period, and treat Reserved Instances and Savings Plans as quasi-capital commitments even though accounting books them as prepaid OpEx.
The Two Models, Defined
CapEx: own the asset, spread the cost
Capital expenditure is spending that creates an asset expected to deliver value for more than one year: servers, storage arrays, network equipment, the building they sit in. The full price is paid up front and recorded on the balance sheet as an asset rather than an expense. Each year a portion of that value is recognized as depreciation on the income statement, typically over three to five years for IT hardware. The economic consequences follow from that structure: large cash outlays at refresh time, a smoother income statement because the expense is spread, capacity that must be sized for peak demand years in advance, and ownership of equipment that ages whether or not it is used.
OpEx: rent the service, expense as you go
Operating expenditure is spending on the ongoing running of the business, recognized as an expense in the period it occurs. Cloud compute, storage, and services billed monthly are OpEx: there is no asset on your balance sheet, no depreciation schedule, and no residual equipment. The economic consequences again follow: no upfront cash, a cost that scales with usage in both directions, capacity that can be added or removed in minutes, and an expense that hits the income statement immediately. For how that flexibility works mechanically, see our explainer on cloud elasticity.
Side-by-Side Comparison
| Dimension | CapEx (on-premises) | OpEx (cloud) |
|---|---|---|
| What you pay for | Equipment you own | Capacity and services you consume |
| Timing of cash | Large outlay up front, at purchase and refresh | Monthly, as used |
| Income statement | Depreciation spread over 3 to 5 years | Full expense in the period incurred |
| Balance sheet | Asset, declining with depreciation | No asset (prepaid commitments excepted) |
| EBITDA effect | Depreciation sits below EBITDA, so EBITDA looks stronger | Expense sits in operating costs, reducing EBITDA |
| Capacity planning | Sized for peak years ahead; idle most of the time | Sized to demand; scales up and down |
| Utilization | Often 20 to 40 percent on average | Pay only for what runs |
| Flexibility | Low; equipment is sunk cost | High; minutes to change |
| Risk profile | Over- or under-provisioning; technology obsolescence | Variable bills; vendor lock-in; cost sprawl |
| Decision owner | Finance, with long approval cycles | Engineering, often without approval |
| US tax treatment | Depreciation; 100 percent bonus depreciation currently available | Deductible as incurred |
The Formulas for a Fair Comparison
Most CapEx-versus-OpEx comparisons fail because they compare a purchase price to a monthly bill. The fair comparison converts both to annual total cost of ownership over the same period, per unit of work delivered.
Annual on-premises TCO equals the hardware purchase price divided by its useful life, plus annual facilities (space, power, cooling), plus maintenance and support contracts, plus the fully loaded cost of the staff who operate it, plus a share of the network and security infrastructure, plus an allowance for refresh and end-of-life. Annual cloud TCO equals the cloud bill, plus the fully loaded cost of the (usually smaller) team that runs it, plus third-party tooling, plus any egress or migration costs. Neither side is just the obvious number.
The adjustment most comparisons miss is utilization. On-premises capacity is bought for peak and typically runs at 20 to 40 percent average utilization, so its true cost per unit of work is the TCO divided by the work actually done, not the work it could do. Cloud, sized to demand, runs closer to full utilization of what you pay for. A fair comparison therefore computes cost per unit of output, such as cost per customer, per transaction, or per thousand requests, on both sides. The break-even is the utilization at which on-premises cost per unit equals cloud cost per unit; above it, owning wins; below it, renting does.
For multi-year decisions, discount both cash flow streams to present value, because the on-premises outlay is front-loaded and the cloud stream is spread. We cover the full method in our guide to cloud total cost of ownership.
A Worked US Example
A mid-size SaaS company needs the equivalent of 40 general-purpose servers. On-premises, that is roughly $600,000 of hardware with a five-year life ($120,000 a year), about $90,000 a year in colocation space, power, and cooling, $60,000 in maintenance contracts, and two infrastructure engineers at a fully loaded $400,000. Annual TCO: about $670,000. The hardware is sized for a peak that occurs a few hours a day; average utilization is 30 percent.
In the cloud, the same peak capacity on demand would cost about $35,000 a month, or $420,000 a year, but because capacity scales to demand the average bill is closer to $18,000 a month, or $216,000 a year. Add one platform engineer at $200,000 and $30,000 in tooling. Annual TCO: about $446,000, roughly a third less than on-premises, and with no $600,000 refresh in year five.
Now change one variable. Suppose the workload is steady, running near peak around the clock at 85 percent utilization. The cloud bill approaches the $420,000 on-demand figure, and even a three-year Savings Plan at a 60 percent discount only brings it to about $170,000 plus staff and tooling, around $400,000 total. On-premises stays at $670,000 but is now delivering nearly three times the work per dollar it was at 30 percent utilization, and with 100 percent bonus depreciation the $600,000 purchase is deductible in year one under current US tax law, which materially improves its after-tax cash position. The comparison is now close, and for the most stable workloads it tips toward owning. The lesson is not that cloud is cheaper or dearer; it is that utilization decides, which is why hybrid estates keep steady baseload on owned or committed capacity and put variable demand in the cloud.
The Nuance: Cloud Is Not Purely OpEx
Three things complicate the clean story, and finance teams should know all three.
- Commitments are quasi-CapEx. An all-upfront three-year Reserved Instance or Savings Plan is recorded as a prepaid expense and amortized monthly, so it stays OpEx in the accounts. Economically it is a capital decision: cash out now, capacity locked for years, no refund if needs change. Govern commitments with the same rigor as a hardware purchase, with a named owner, a utilization target, and a review cadence. Our guide to
Reserved Instances versus Savings Plans versus Spot covers how to size them.
- Some cloud costs can be capitalized. Under US GAAP, ASU 2018-15 allows certain implementation costs of a cloud computing arrangement, such as configuration and customization during setup, to be capitalized and amortized over the arrangement's term, while the service fees themselves remain OpEx. This can matter for large migrations; confirm the specifics with your auditors.
- The decision moved. On-premises CapEx required a purchase order and a finance review. Cloud OpEx is incurred by an engineer launching an instance. The shift to OpEx is also a shift in who decides, which is why cloud spend sprawls in ways hardware never could, and why the FinOps discipline exists. Our guide to
what FinOps is explains how organizations put governance back around spend that no longer passes through procurement.
Why This Matters More for SaaS
For a software company, cloud infrastructure that delivers the product sits in cost of revenue, so every dollar of cloud OpEx reduces gross margin directly and visibly. A company at $10 million in annual revenue spending $2 million on cloud has given up 20 points of gross margin to infrastructure before any other cost. Investors benchmark SaaS gross margins in the 70 to 80 percent range, and the difference between a company that scales revenue faster than its cloud bill and one that does not is the difference between margin expansion and margin compression. That is why SaaS finance teams track cloud cost per customer, per tenant, and per unit of usage rather than the total: it is the number that connects infrastructure decisions to the metrics the board watches, the subject of our guides to AI gross margin and SaaS profitability and the FinOps KPIs that show cost control is working.
How to Decide, Workload by Workload
- Favor cloud OpEx when demand is variable, growing, or uncertain; the company is early and cash matters more than margin; speed of change matters; or the workload is new and its shape is unknown.
- Favor owned or committed capacity when demand is steady and predictable at high utilization; the workload will run unchanged for years; data gravity or compliance favors a fixed location; or tax treatment of depreciation materially improves the after-tax case.
- Favor a mix when the estate has both, which is nearly always. Steady baseload on commitments or owned hardware, variable demand on pay-as-you-go, and a single view of cost per unit across both so the split can be rebalanced as utilization changes.
The last point is where this stops being a finance question and becomes an operating one. Whichever model a workload sits on, the decision only stays right if someone can see cost per unit continuously and move workloads when the numbers change. That continuous view, across cloud, commitments, and whatever remains on-premises, is what Opslyft's cost visibility provides, and it is the input every formula in this guide depends on.
The Bottom Line
CapEx and OpEx are not better or worse; they are different ways of paying for the same capacity, with different effects on cash, margins, flexibility, and who decides. Cloud moved infrastructure decisively toward OpEx, and for most variable workloads that is cheaper and faster. For steady, high-utilization workloads the case is closer than the cloud industry suggests, especially with current US depreciation rules, and commitments like Reserved Instances bring some of CapEx's lock-in back into the cloud bill. Compare the two on total cost per unit of work over the same period, treat commitments as the capital decisions they are, measure cloud cost in the units your margin is measured in, and expect the right answer to differ by workload. Then confirm the accounting with your own finance team, because the treatments here are general and your books are specific.
FAQs
CapEx (capital expenditure) is money spent up front on assets you own, such as servers and data center equipment, recorded on the balance sheet and depreciated over several years. OpEx (operating expenditure) is money spent on an ongoing basis for services you use, such as cloud compute billed monthly, expensed as incurred. Cloud shifts most infrastructure spend from CapEx to OpEx, trading ownership and depreciation for flexibility and a recurring bill.
Mostly OpEx. Pay-as-you-go cloud charges are operating expenses recognized in the period they occur. Two nuances: upfront payments for Reserved Instances or Savings Plans are recorded as prepaid assets and expensed over the term, and under US GAAP (ASU 2018-15) certain implementation costs of a cloud arrangement can be capitalized and amortized. The cloud service itself is never a capital asset you own.
It depends on utilization. Cloud is cheaper for variable, growing, or uncertain workloads because you pay only for what you use and avoid capacity bought for peak. On-premises can be cheaper per unit for steady, predictable, high-utilization workloads over a full hardware lifecycle, especially with 100 percent bonus depreciation available under current US tax law. The honest comparison is total cost of ownership over three to five years, not sticker price.
Convert both to an annual total cost of ownership over the same period. For CapEx, add the purchase price divided by useful life to the annual running costs: facilities, power, cooling, maintenance, staff, and refresh. For OpEx, add the cloud bill to the staff and tooling needed to run it. Then compare on a per-unit basis, such as cost per customer or per transaction, and adjust for utilization, since on-premises capacity bought for peak sits idle most of the time.
Because the two hit financial statements differently. CapEx appears as depreciation over years, smoothing the income statement and preserving EBITDA; OpEx hits the income statement immediately and reduces EBITDA in the period. CapEx consumes cash up front; OpEx preserves cash but commits future periods. Investors in SaaS companies also watch gross margin, and cloud OpEx that scales with usage sits in cost of revenue, directly affecting it
Not in accounting terms, but they behave like it economically. An all-upfront three-year Reserved Instance is a prepaid expense on the balance sheet, amortized monthly as OpEx, so it never becomes a depreciable capital asset. But it commits cash up front and locks in capacity for years, which is exactly the trade-off CapEx involves. Treat commitments as quasi-capital decisions and govern them accordingly.