Micro SaaS Is Back, and the One-Person Version Has a Ceiling Nobody Prices In

Micro SaaS is genuinely back. The cost of building a narrow software product has collapsed, and solo operators are shipping profitably. The economics still cap out early. Closed micro SaaS deals change hands at a 3.9x median profit multiple, and the costs that decide whether you get there never fell at all.

What "micro SaaS" actually means in 2026

A micro SaaS is a narrow subscription software product serving one workflow for one audience, operated by a team of one to about five people. It has no official definition, no industry classification code, and no audited statistics anywhere.

That absence is worth sitting with before you read another benchmark. Every median revenue figure circulating for this category comes from a self-selected survey, a public revenue dashboard, or a marketplace listing page. None of them is an audited account, and all of them are drawn from operators who chose to be visible.

My position is that you should treat every published micro SaaS median as unusable for planning. Use them to understand the shape of the category. Do not use them to forecast your own revenue, because the sample is defined by success in a way that guarantees the number flatters you.

The base rate nobody quotes

One-person businesses are not a new phenomenon and they are not rare. The US Census Bureau counted 30,427,808 nonemployer establishments in 2023, up by more than 616,000 on the prior year.

Those businesses generated close to $1.8 trillion in receipts, equal to 6.4% of US GDP, and they made up 78.4% of all US establishments. The default American business already has no employees. It has for decades.

What changed is not the existence of one-person firms. It is the arrival of one-person firms selling recurring software revenue, which historically demanded a team for support, security and sales. That is the specific claim worth testing, and it is narrower than the way the category usually gets sold.

The honest limitation here is that Census does not publish a clean count of one-person software businesses. Its software categories mix contract development, freelance consulting and product revenue into one receipts line. Nobody can tell you how many real micro SaaS companies exist, including me.

What AI collapsed, and what it left alone

Building got cheaper than deciding what to build

Adoption is now effectively universal among people who write software. Google's 2025 DORA research, based on nearly 5,000 technology professionals, found that 90% of respondents use AI at work. Stack Overflow's 2025 survey put the figure at 84% using or planning to use AI tools.

Read that as a warning rather than an opportunity. When a capability reaches 90% adoption it stops being an advantage and becomes a cost of entry. Everyone building a competing narrow tool is building it at the same reduced cost you are.

The scarce input moved upstream, to knowing which narrow workflow is worth automating and having access to the buyers who feel that pain. Neither of those got cheaper. Both got more crowded, which is the argument made at more length in the piece on what actually makes a thin AI product defensible.

Maintenance did not follow the build cost down

The most careful measurement available points the other way. METR ran a randomised controlled trial with 16 experienced open-source developers across 246 real tasks, in repositories they had worked in for roughly 5 years. Developers using AI tools were 19% slower, while estimating afterwards that they had been 20% faster.

Stack Overflow's respondents describe the mechanism. 66% name almost-right solutions as their top frustration, and 45.2% say debugging AI-generated code takes more time than writing it themselves. Only 3.1% highly trust the accuracy of what these tools produce.

DORA found the same tension in delivery data, reporting a positive relationship between AI adoption and throughput alongside a negative relationship with delivery stability. You ship more, and more of what you ship breaks. For a solo operator that trade lands entirely on one calendar.

Two cautions on that chart, because it is the single most over-quoted result in this debate. It covers early-2025 tooling, and it studies mature codebases with 5 years of accumulated context. A micro SaaS operator working in a small codebase written last month is in a very different setting, and I return to that in the section on where this argument fails.

The unit economics of a one-person portfolio

Revenue per operator is the metric that matters

Acquire.com publishes the closest thing this category has to transaction data. Its January 2026 report covers confirmed closed SaaS acquisitions below $10 million, drawn from anonymised platform records rather than a survey.

Three figures define the shape of the business. The median profit multiple was 3.9x in both 2024 and 2025. The average profit margin on those businesses was 71%, up from 67% in 2023. Average time on market was 81 days.

Work the arithmetic through and the ceiling appears immediately. A product doing $8,000 in monthly recurring revenue at that 71% margin throws off about $68,000 of annual profit. At the median multiple it sells for roughly $266,000. That is a good year and a decent car, not an exit.

The margin is real, the base is small

Micro SaaS margins look like enterprise software because the cost structure is genuinely light. There is no sales team, no office, and increasingly no contract engineering. The margin is the strongest thing about the model and it is not in dispute.

The weakness is that a high percentage of a small number is still a small number. Multiples here apply to profit, not to revenue, which is the opposite of how venture-backed software is valued. That difference alone explains most of the confusion when indie builders compare their outcomes to funded startups.

One survivorship caveat belongs on every figure in this section. Acquire.com reports closed transactions, so it describes products that found a buyer. Everything that listed and never sold, and everything that quietly stopped, is absent from that dataset by construction.

What the celebrated exits actually show

Base44 was not one person at the exit

Base44 is the case everyone cites, and the details cut against the headline. Wix announced the acquisition on 18 June 2025 for about $80 million in initial consideration, with earn-out payments running through 2029.

Buried in the same announcement is the tell. Wix expected to pay roughly $25 million in retention bonuses to Base44 employees during 2025. A company with one person in it has nobody to retain.

Israeli outlet Calcalist reported the company was six months old with 8 employees at the point of sale, founded by Maor Shlomo, who had previously co-founded a company that raised $125 million. Solo at launch is a real and impressive thing. Solo at an $80 million exit did not happen here.

OpenClaw shows the other failure mode

The second case cuts the same way from a different direction. Peter Steinberger launched OpenClaw in November 2025 as a single maintainer, and it reached extraordinary adoption within weeks.

By 15 February 2026 Steinberger had joined OpenAI, with the project moving into a foundation that OpenAI agreed to support. Read that structure carefully. Viral distribution arrived, and within three months the operating burden was transferred to an organisation and a sponsor.

Neither outcome is a failure. Both are evidence that the one-person structure is a starting configuration rather than a durable one, and that the transition happens fast when demand shows up.

The third row is the one to plan against. The first two rows are the two ways a genuinely successful solo product stops being a solo product.

The four costs that did not fall

Every cost line in a software business moved between 2023 and 2026, but they did not move together. Sorting them is the whole planning exercise for a solo operator.

Support scales with customers, not with code. A model can draft a reply in seconds. It cannot decide whether to issue a refund, and it cannot carry the accountability when an answer is wrong. Every additional 100 paying customers adds a fixed weekly load that no tool removes.

Trust and compliance are purchased, not written. SOC 2 audits, data processing agreements, penetration tests and liability insurance cost money and calendar time. This is the hard boundary between selling to individuals and selling to companies, and it is where most solo products stop.

Distribution never got cheaper. Cheaper building raised the supply of narrow tools while attention stayed fixed. That is a straightforward squeeze on the same channels, and it interacts badly with platform owners absorbing adjacent features, a pattern covered in the analysis of point tools being absorbed into agent ecosystems.

Key-person risk is priced by every buyer. If the product cannot run for a fortnight without you, an acquirer is buying a job rather than an asset. That discount is applied quietly and it is applied every time.

Directions in the middle column are my assessment of the cited evidence, not measured cost data. No public dataset tracks micro SaaS cost lines over time, which is a real gap.

Where this argument is weakest

The case above leans on two studies and one marketplace dataset. Here is where each one strains.

The case that the ceiling keeps rising

Every constraint I listed is a candidate for automation rather than a law of physics. Tier-one support handled by an agent, compliance evidence collected automatically, and onboarding driven by the product itself would each move the ceiling directly upward.

Base44 reached an $80 million outcome in roughly six months, which is not the behaviour of a category with a hard cap. The most senior people in the industry now talk openly about a one-person billion-dollar company arriving, and while that framing is promotional, the direction of tooling supports it more than it contradicts it.

What the sceptics get wrong

The METR result is quoted far more loosely than it deserves. It measured 16 developers on mature repositories they had worked in for about 5 years, using tools available between February and June 2025. Those are precisely the conditions where AI assistance helps least.

A micro SaaS operator usually works in a small codebase they wrote recently, with no legacy constraints and low review overhead. Importing a 19% slowdown into that setting overstates the case, and I would not do it. The honest reading is that the productivity gain is real and concentrated in new code, then decays as the codebase ages.

Nothing here can be settled with current data. There is no audited dataset of one-person software businesses, no reliable count of how many exist, and no failure-rate series. Until a statistical agency or a payment processor publishes one, everyone in this debate is arguing from anecdotes, including this post.

The ceiling, priced honestly

Given all of that, here is where I think the model works and where it stops. These bands are my judgement from the evidence above, not measured thresholds, and you should treat them as a starting hypothesis to test against your own numbers.

Below roughly $2,000 in monthly recurring revenue, a micro SaaS is a side project with real cash flow. It survives on the operator's spare capacity, and most products stay here permanently.

Between roughly $2,000 and $20,000 monthly, the model does what it promises. At a 71% margin that band produces a genuine professional income for one person, and the support load stays inside one calendar. This is the target that matches the evidence, and it is far less exciting than the case studies.

Above roughly $20,000 monthly, the four costs above start demanding other people. Enterprise buyers appear with security questionnaires, support volume outgrows the mornings, and the choice narrows to hiring, selling, or capping growth deliberately. Capping deliberately is a legitimate answer that nobody writes case studies about.

If you are choosing between this and a funded startup, the trade is legible. A micro SaaS pays you sooner, values you at a multiple of profit, and hands you the whole decision. The comparison to venture-scale outcomes only makes sense if you were ever going to raise, which is the same question examined in the test for whether a startup sits inside a platform's kill zone.

One cost line deserves specific attention before you model anything. If your product calls a model on every user action, your gross margin is exposed to inference pricing in a way traditional software never was, and that exposure is examined in the breakdown of what inference costs do to software margins.

Frequently asked questions

Where to start this week

Two things, and neither involves writing code.

First, log every hour you spend on support, compliance and admin for one week, separately from building. That ratio is the single number that predicts your ceiling, and almost nobody measures it before they hit the wall.

Second, write down what your product would sell for today using the median multiple above. Take your last 12 months of profit and multiply by 3.9. If the answer disappoints you, that tells you whether you are building an income or an asset, and those need different decisions from here.

Related reading