SaaS Gross Margin Benchmarks: What Good Looks Like

What is a good SaaS gross margin in 2026?

The median software gross margin published for 2026 is 80 percent, and the median gross margin on total revenue for the same companies is 76 percent, so the first question is not where you sit in the band but which band you belong in.

Both figures come from the 2026 SaaS & AI Performance Benchmarks, published by Aleph and Benchmarkit on 1 June 2026, covering 342 B2B SaaS and AI-native companies on full-year 2025 actuals. Those are participant-reported figures from management accounts, not audited filings, which matters for every comparison you make against them. The top quartile runs at 86 percent and above, the bottom quartile at 50.

The two calculations, stated once so the rest of this page can refer to them:

Software gross margin = (subscription revenue minus subscription cost of revenue) divided by subscription revenue.

Total gross margin = (total revenue minus total cost of revenue) divided by total revenue.

Those two medians are 4 points apart, and that distance is the first thing to understand about any gross margin benchmark. It is not a disagreement between publishers. It is the price of a definition, and it is smaller than the distance most companies discover between their reported margin and their real one.

SaaS gross margin benchmarks by ARR band and pricing model

Gross margin rises with scale and falls with usage-based pricing, and both movements are visible in the 2026 Aleph and Benchmarkit dataset. Read the sub $5M line as the seed and early Series A band, and the $50M to $100M line as what the same companies reach several rounds later.

All figures: Aleph and Benchmarkit, 2026 SaaS & AI Performance Benchmarks, published 1 June 2026, full-year 2025 actuals.

Two readings matter more than the numbers themselves. A company below $5M ARR sitting at 72 percent is at its own median, not 8 points below the market, because the 80 percent figure pools companies of every size, and smaller companies run lower in the same report. And a usage-priced company at 62 percent is at its own median too: usage revenue carries infrastructure cost that scales with consumption, which is a business model rather than a failure of discipline.

Software gross margin vs total gross margin: which one you are being compared to

Software gross margin covers subscription revenue and the cost of serving it. Total gross margin adds implementation, onboarding and training revenue along with the people who deliver them, and those people cost far more per dollar of revenue than infrastructure does.

Public filers report the two separately, which is the clearest available evidence that the distinction is not a benchmarking subtlety.

Margins here are computed from the revenue and cost of revenue lines as filed under GAAP. Both companies also publish non-GAAP margins that exclude stock-based compensation and the amortisation of acquired intangibles, and those run higher. That is a third basis, which is the point of this section: a margin figure means nothing until you know which basis produced it.

The size of the gap follows the size of the non-subscription line. ServiceNow reported $110M of professional services revenue against $139M of cost, a gross loss of $29M in the quarter, which is the whole of the gap between its two columns above. GitLab’s second line is mostly licence rather than services, with professional services at $6.7M or under 3 percent of revenue, so its blended margin moves barely 1 point.

The lesson is not that services are bad business. It is that a blended number carries a drag the software benchmark does not measure, and that the drag is proportional to a mix you know and your reader does not. If services are more than a rounding error in your revenue, produce both margins every month. If they are not, say so, because a reader comparing you to the 80 percent figure will otherwise assume you have quietly folded them in.

Why your reported gross margin may be higher than your real one

In 17 of 22 engagements with SaaS and AI companies, hosting, compute or support costs sat in operating expenses instead of cost of revenue, and correcting the classification moved reported gross margin by 3 to 13 percentage points, with a median of 7. The 22 are every engagement between September 2025 and August 2026 where we took over books previously kept by someone else, in-house or outsourced.

That distribution matters more than the headline. A 3-point correction changes a sentence in a board deck. A 13-point correction changes what the business is: a company reporting 82 percent that turns out to run at 69 is not a high-margin software company having a slow quarter, it is a services-heavy or infrastructure-heavy company that has never seen its own economics.

The mechanism is always the same. Costs enter the ledger where the invoice suggests, not where the economics put them. A cloud bill looks like a utility, so it lands in operating expenses next to the rent. Support salaries look like payroll, so they land with the rest of payroll. Nobody decides to overstate gross margin. The chart of accounts decides it, once, and then repeats the decision every month.

This is accounting accuracy rather than strategy, which is why it belongs to controller work rather than to a forecast. A forecast built on a margin that is 7 points wrong is not a cautious forecast, it is a confident one pointing in the wrong direction.

Why the gross margin correction is larger for AI companies

All 9 AI companies in those 22 Offset Partners engagements carried misclassified direct costs, with corrections of 6 to 13 percentage points, against 8 of the 13 SaaS companies and corrections of 3 to 7 points. The two ranges barely overlap.

The reason is structural rather than cultural. A SaaS company misfiles hosting, which is a real cost but a stable and comparatively small one. An AI company misfiles inference, which scales with every unit of usage and is frequently the largest single line in the cost of serving a customer. The same accounting habit produces a much larger error because the underlying number is much larger. How to model and categorise those costs is the subject of our AI compute cost guide. What compute does to the level of an AI company’s margin, and how much of revenue it takes, is on our AI startup gross margin benchmarks page.

Set that against what the benchmark reports. Aleph and Benchmarkit publish an affirmative finding, not a hedge: software gross margin held between 79 and 81 percent across four years, and, in their words, “AI infrastructure costs have not yet compressed software margins at the median”, with the report itself described as unable to prove an AI margin compression hypothesis on the data it had.

Our reading, offered as a reading and not as a measurement: a benchmark built on participant-reported margins measures reporting as well as economics. If compute sits in operating expenses for a meaningful share of participants, a stable median is exactly what you would expect to see whether or not the underlying economics have moved. In the books we have opened, an AI company’s reported margin was the least reliable number on the page.

The founder-level consequence is simpler than the market-level question. If your margin looks like a software company’s margin and your product runs inference on every request, that is a reason to check the classification, not a reason to relax.

What belongs in SaaS cost of revenue (COGS), and what does not

Cost of revenue holds what it takes to deliver what existing customers have already bought. The test is what the cost is for: if it delivers that, it belongs in cost of revenue; if it is spent to win, renew, expand, build or run the company, it is an operating expense.

Three lines cause almost all of the argument. Fully burdened support means salary, benefits, payroll taxes and the tools that team uses, not base salary alone. Customer success has to be split rather than assigned whole, because the same team often does delivery and renewal, and putting all of it in either place is wrong in a predictable direction. And model training is an operating expense while inference behind a paid feature is not, which is the distinction that most often gets collapsed.

This split is our operating convention, aligned with how public filers present subscription cost of revenue separately from services cost of revenue. It is not a rule handed down by an accounting standard, and a different auditor may draw one of these lines elsewhere. What matters more than the edge cases is that you draw it once, write it down, and stop moving it.

How to recompute your SaaS gross margin on the benchmark’s basis

Four steps produce the two numbers a benchmark can actually be compared to.

1. Pull every cost line in operating expenses and test each one. Does this cost deliver what existing customers have already bought? Everything answering yes moves to cost of revenue.
2. Burden the people you moved. A support salary moved at base cost, without benefits, payroll taxes and the tools that team uses, understates the correction. Use the same burden rate you use everywhere else, and use it consistently.
3. Split revenue and cost between subscription and services. Subscription revenue with its infrastructure, APIs and support. Services revenue with its delivery labour.
4. Compute both margins. Software margin on subscription only, total margin on everything.

A worked example. A company bills $250,000 a month, $210,000 of subscription and $40,000 of implementation, so $2.52M of ARR against a $3M total revenue run rate. Its profit and loss shows $40,000 of cost of revenue, so it reports an 84 percent gross margin.

Step 1 finds $9,500 of production hosting and $8,000 of burdened tier 1 and tier 2 support in operating expenses. Cost of revenue becomes $57,500, and total gross margin is $192,500 on $250,000, or 77 percent. The reported figure was 7 points high, which is the median correction on our engagements.

Step 3 splits it. Of the original $40,000, $18,000 was third-party APIs serving the subscription and $22,000 was implementation labour. Subscription cost of revenue is therefore $18,000 plus $9,500 plus $8,000, or $35,500 against $210,000 of subscription revenue: an 83 percent software margin. Services run $22,000 against $40,000, a 45 percent margin.

The company that reported 84 percent is at 83 on software and 77 on total revenue. It had been comparing 84 against the all-sizes median of 80 and reading a 4-point lead. Against the band that actually applies to it, 72 percent for companies below $5M ARR, its software margin of 83 sits 11 points above the median. Its position is stronger than it believed, on a number it can now defend. Its total margin sits just above the all-sizes median of 76, and the report publishes no total margin band by ARR, so that second comparison is the looser of the two.

One honest caveat on both comparisons, and it follows from everything above. The benchmark is built on what participants reported, and this company has just restated its own figure. A corrected number read against an uncorrected population will tend to read low rather than high, so treat the gap as a floor. That is an argument for knowing your own basis, not against comparing at all.

Nothing about the business changed. Two things changed for the reader. The comparison became valid, which is the only reason any of these numbers mean anything. And the services line surfaced at 45 percent, which is now the visible decision: price it, scope it, or accept that it costs 6 points of blended margin.

The check that tells you the work is done. Can you produce subscription gross margin and total gross margin for last month from your ledger, without a spreadsheet on the side, and would both survive an investor restating them? If not, the reclassification is not finished, and any comparison to a published band is a comparison of two different quantities.

What actually moves SaaS gross margin, and what does not

Once the number is correct, four things move it by points rather than basis points: pricing and packaging, infrastructure unit cost, the support model, and the share of services in revenue.

Pricing moves it fastest, because a price increase carries almost no incremental cost. Infrastructure unit cost moves next, through committed-use discounts, rightsizing and, for AI products, cost per inference call. The support model moves it through deflection and self-serve rather than headcount cuts, since a team that shrinks while ticket volume holds produces a margin gain and a retention loss. Services move it by being priced at or above cost, or by being scoped out of the offer entirely, and ServiceNow’s gross loss on professional services shows a company at scale choosing neither.

What does not move it, despite appearing on most lists: renegotiating the software your team uses internally, freezing engineering headcount, and cutting marketing. Those are operating expenses. They improve profitability, which is a different objective on a different line, and a founder who attacks them expecting gross margin to move will be disappointed twice. The ranking in this section is our judgment from client work rather than a measured result.

When a gross margin correction changes a decision

A 7-point correction is not a reporting adjustment. It changes three decisions that were already made on the old number.

Runway. This month’s burn does not change: the misfiled costs were being paid either way, only on a different line. What changes is the forecast. A plan that applies the reported margin to every new dollar of revenue, while the misfiled costs sit in operating expenses as if they were fixed, books 7 cents too much gross profit on each dollar of growth, and that error compounds into the months of runway the plan shows. The SaaS runway calculator shows what the corrected margin does to the number of months.

Pricing and hiring. A company that believes it earns 84 cents on the marginal dollar will hire and discount differently than one earning 77. Both decisions were taken with the wrong input, and both are expensive to reverse.

Diligence. An investor will restate your gross margin on their own definition, and they will do it after the term sheet rather than before. Learning that the number was never yours, at that moment, costs more than the margin itself: it puts every other number in the model into question, which is how a confirmatory diligence turns into a real one.

How Offset Partners handles gross margin classification

We work only with SaaS and AI companies, so hosting, inference, burdened support and services delivery costs are routine classification work here rather than something discovered on your time and billed as a surprise.

The diagnostic starts with the classification rather than the forecast, because that is the order the numbers require. Controller support starts from $2,500 per month and fractional CFO support from $3,500, and which one you need depends on what the diagnostic finds: if the margin is wrong, the answer is the accounting layer, and buying CFO time to fix a chart of accounts is the most expensive available way to fix a chart of accounts.

We will also tell you when the answer is no. If your ledger already produces both margins and they hold up when restated, there is nothing here for us to sell you, and we would rather say so.

Sources

External figures on this page, with the date each was published.

• Median software gross margin of 80 percent, median total revenue gross margin of 76 percent, quartiles at 86 and 50 percent, the sub $5M and $50M to $100M ARR cuts, the 62 percent usage-only figure, the 79 to 81 percent four-year stability, and the quoted finding that AI infrastructure costs have not yet compressed software margins at the median: 2026 SaaS & AI Performance Benchmarks, Aleph and Benchmarkit, published 1 June 2026. Population: 342 B2B SaaS and AI-native companies on full-year 2025 actuals, participant-reported from management accounts, with gross margin cuts drawn from 228 to 232 of them and no participant count published for the individual ARR cuts.
• ServiceNow subscription and total gross margin and the professional services gross loss: ServiceNow Form 10-Q for the quarter ended 30 June 2026. Margins computed from the revenue and cost of revenue lines as filed under GAAP, a different basis from the survey medians above.
• GitLab subscription and total gross margin, and the split of its non-subscription line into $18.2M of licence and $6.7M of professional services and other: GitLab Form 10-Q for the quarter ended 30 April 2026, revenue disaggregation note. Margins computed from the revenue and cost of revenue lines as filed under GAAP.
• The cost of revenue split in the table above is Offset Partners’ operating convention, aligned with the subscription and services presentation used by the filers cited, not a requirement of an accounting standard.

The classification finding, 17 of 22 engagements and a 3 to 13 percentage point gross margin correction, median 7, and the split between 9 of 9 AI companies at 6 to 13 points and 8 of 13 SaaS companies at 3 to 7 points, is Offset Partners’ own data. Population: engagements with SaaS and AI companies between September 2025 and August 2026 where we took over books previously kept by someone else, in-house or outsourced.

Related resources

• SaaS runway calculator
• AI startup gross margin benchmarks
• AI compute costs: a finance guide for founders
• SaaS chart of accounts template
• SaaS finance readiness checklist
• When to hire a fractional CFO

Related services

• SaaS controller services
• Fractional CFO services
• SaaS bookkeeping and monthly accounting
• Book a SaaS finance diagnostic