# The economics of a one-person AI business: what the MRR screenshots leave out > Source: > Published: 2026-06-21 00:00:00+00:00 The Ledger · Income # The economics of a one-person AI business: what the MRR screenshots leave out - AI product gross margin, 2026 (projected) - ~52%, vs 70-80% for mature SaaS (ICONIQ) - Monthly churn, products under $25/mo - 6.1% median, about half the base in a year (ChartMogul) - Inference price for a fixed capability - falling ~50x per year (Epoch AI) - Indie products making $0 - 54%+, only ~5% clear $100k/year (Scraping Fish, 2022) The MRR screenshot is the most shared number in indie AI and the least useful. What the research says about gross margins, churn, fees, and what a solo founder actually keeps. Open any indie maker feed and you find the same image: a revenue dashboard, a green line going up, an MRR number circled in red. It is the most shared number in indie AI, and the least useful one. Revenue is what a customer pays. Income is what you keep after the model bill, the card fees, the refunds, the tax you are holding, and the customers who quietly leave. Those are very different numbers, and the gap between them is wider for an AI product than for almost anything else you could build. Here is what the research actually says about the economics of a one-person AI business, and where the screenshots stop telling the truth. ## Almost nobody gets the screenshot Start with the survivor problem, because it shapes everything else. When Scraping Fish pulled every Indie Hackers product with Stripe-verified revenue in 2022, all 937 of them, [more than half were making nothing at all, and only about 5% cleared roughly $100,000 a year](https://scrapingfish.com/blog/indie-hackers-revenue), a level its author notes is “not that hard to earn as a software engineer in a full time job.” The conclusion was blunt: “success in the world of indie developers is an outlier business.” Now compare that to the rooms where the screenshots come from. At MicroConf in 2025, a bootstrapper conference, founder Rob Walling reported that [28% of attendees were doing more than $100,000 in monthly recurring revenue](https://www.linkedin.com/posts/robwalling_the-state-of-bootstrapped-saas-in-2025-is-activity-7315372913732329473-obpo). That is not a contradiction, it is selection: one number is the whole population, the other is the people who already won and bought a conference ticket. Watch the units too. MicroConf’s 28% is $100k a month. The Indie Hackers 5% is $100k a year. The base rates are sobering even outside software. U.S. government data finds [about 34.7% of business establishments born in 2013 were still operating ten years later](https://www.bls.gov/opub/ted/2024/34-7-percent-of-business-establishments-born-in-2013-were-still-operating-in-2023.htm), and roughly half survive five years. AI has made the first dollar faster: [Stripe Atlas reports its 2025 cohort reached first payment in a median of 34 days, down from 38, with 20% charging a customer inside 30 days versus 8% in 2020](https://stripe.com/blog/stripe-atlas-startups-in-2025-year-in-review). But getting to the first dollar quickly is not the same as reaching a number worth screenshotting. ## The margin you do not have Here is the part most pricing advice skips: an AI product is not an 80-percent-gross-margin SaaS business, and pretending it is will quietly bankrupt you. Classic software is cheap to serve. Bessemer’s cloud benchmarks put [the best SaaS gross margins at 80% and above](https://www.bvp.com/atlas/scaling-to-100-million), with a typical cloud business around 65 to 70%. AI is structurally lower. Andreessen Horowitz flagged this early. Its 2020 essay on [the new business of AI](https://a16z.com/the-new-business-of-ai-and-how-its-different-from-traditional-software/) pegged AI gross margins “often in the 50-60% range,” dragged down by the “25% or more of revenue” that goes to cloud and compute. Six years on, the gap has narrowed but not closed: ICONIQ’s [2026 State of AI snapshot](https://www.iconiq.com/growth/reports/2026-state-of-ai-bi-annual-snapshot) reports average AI product gross margins of 41% in 2024, 45% in 2025, and a self-projected 52% in 2026, still well under the 70 to 80% a mature SaaS business takes for granted. The reason is simple and it does not go away: your cost of goods is metered. Every query a user runs costs you tokens. SaaS hosting amortizes toward zero as you scale; inference does the opposite. ICONIQ finds [model inference rising from 20% of AI product spend before launch to 23% at the scaling stage](https://www.iconiq.com/growth/reports/2026-state-of-ai-bi-annual-snapshot), becoming, in its words, “the dominant cost driver at scale.” Bessemer’s 2025 taxonomy makes the danger concrete: its usage-heavy “Supernova” companies run at [about 25% gross margin, often negative](https://www.bvp.com/atlas/the-state-of-ai-2025), against roughly 60% for the more disciplined “Shooting Stars.” This is the trap a flat monthly fee walks straight into. If you charge $20 a month and a power user burns $25 of tokens, you are paying them to use your product. a16z’s own partners note that [the heaviest sliver of users drives a wildly disproportionate share of cost](https://a16z.com/questioning-margins-is-a-boring-cliche/), and that rate limits on the top 5% cut spend “with limited revenue impact.” A flat fee on a metered cost is only safe if you cap the tail. ## The tailwind: token prices are collapsing The counterforce is real, and it is the best news in this whole piece. The price of a fixed amount of AI capability is falling faster than almost any input cost in business history. a16z calls it [LLMflation](https://a16z.com/llmflation-llm-inference-cost/): the cost to reach GPT-3-level quality fell from $60 per million tokens in late 2021 to $0.06 by late 2024, roughly 10x a year. Independent analysis from Epoch AI, across six benchmarks, [puts the median decline at about 50x per year](https://epoch.ai/data-insights/llm-inference-price-trends), accelerating to roughly 200x per year for data since January 2024. So if your feature set holds still, the same product gets cheaper to serve every quarter. That is genuinely on your side. The catch is the asterisk every careful read includes: those declines are for a fixed capability, and founders almost never hold capability fixed. You add the smarter model, the longer context, the agent loop, and your per-user cost climbs right back up while the screenshot stays the same size. Falling prices help. They do not hand you SaaS margins. ## The leaky bucket nobody photographs Now the part that turns a good month into a bad year: churn, and it is worst exactly where solo products live. Cheap, self-serve products churn hard. ChartMogul’s benchmarks put [median monthly customer churn at 6.1% for products under $25 a month, against 2.2% above $500](https://chartmogul.com/blog/good-customer-churn-rate/). At 6.1% a month, you lose roughly half your customers in a year before you have sold anyone new. The AI-native numbers are worse: ChartMogul’s 2025 retention study found [products priced under $50 a month keeping just 23% of gross revenue and 32% net, about 20 points below comparable SaaS](https://chartmogul.com/reports/saas-retention-the-ai-churn-wave/). The escape hatch that rescues real SaaS, expansion revenue where existing customers pay you more so growth outruns churn, is mostly closed to you down here. Even [the top quartile of products under $25 a month retains only 64.7% of customers](https://chartmogul.com/blog/good-customer-churn-rate/), meaning even the best performers at the bottom of the market lose a third of their base every year. A one-person AI business at a low price is running up a down escalator: every month starts with a hole to refill before any growth counts. ## Revenue minus the cuts the screenshot hides Even the revenue you do collect is not the revenue you keep. Walk the gap MRR never shows: **Card fees.** Stripe’s standard rate is[2.9% plus 30 cents domestically, climbing to as much as 5.4% plus 30 cents on an international card that needs currency conversion](https://stripe.com/pricing). Sell globally, as most digital products do, and a real slice of every sale is gone before anything else touches it.**Refunds.** When you refund a customer,[Stripe returns their money but keeps the original processing fee](https://support.stripe.com/questions/understanding-fees-for-refunded-payments). The sale reverses; the fee does not.**Disputes.** A chargeback costs you[a $15 fee you only get back if you win](https://stripe.com/pricing), on top of the lost sale. On a $20 subscription, one dispute can erase the margin on a dozen other customers.**Payments that silently fail.** Cards expire and decline. Paddle (ProfitWell) estimates[involuntary churn is 20 to 40% of all churn](https://www.paddle.com/resources/reduce-voluntary-and-involuntary-churn), and Baremetrics’ own data points to[around 9% of MRR lost to failed payments](https://baremetrics.com/blog/recover-failed-payments-save-lost-revenue). Without dunning, that is revenue you “earned” that never arrives.**Tax you owe.** Sales tax and VAT ride on top of the price but are a liability, not income. The EU’s average standard VAT is[21.9%](https://taxfoundation.org/data/all/eu/value-added-tax-vat-rates-europe/), and on digital sales a seller outside the EU owes it from the first euro. That money flows through your dashboard and right back out. ## The screenshot, minus everything Put it together and the circled MRR number is doing a lot of lying by omission. Take a $30,000-a-month screenshot at face value, then subtract a 50% gross margin, the card-fee haircut, the slice lost to failed payments and refunds, the tax you are holding for a government, and the fact that you are the 1-in-20 outlier the screenshot never mentions. What you keep is a different, smaller, and far more real number. That is not a reason to skip the AI product. Most of those leaks are levers you control, and the same research that makes the picture sobering also points straight at what grows the number you keep. ## What actually moves your income **Charge more, do not just sell more.** This is the biggest lever in the whole piece, and the data is one-sided. Under $25 a month you churn at 6.1%, and an AI product under $50 keeps just 23% of its revenue; over $250 a month, retention jumps to 70%. A higher price buys you a customer who stays and a margin that survives the token bill. For most solo AI products the move is up-market, not more users. How you land on the number is its own craft: see[pricing psychology that holds up](/ledger/pricing-psychology-that-holds-up/)and[what to charge for AI work](/ledger/pricing-ai-work/).**Recover the income you already earned.** Involuntary churn is 20 to 40% of all churn, and roughly 9% of MRR can fail to collect from expired and declined cards. Dunning, the automated retries and card-update nudges, claws that back with zero new customers. It is the cheapest income you will ever add.**Protect the margin so revenue becomes income.** Cap or meter the heavy tail so a handful of power users cannot turn a profitable plan into a loss, and route each job to the cheapest model that can actually do it. With inference prices falling roughly 50x a year, the model that was too expensive last quarter is often fine now.**Sell where retention lives.** The retention gap is not random: higher-value and business buyers stay far longer than cheap consumer subscriptions. Choosing who you sell to, and positioning for it, is an income decision as much as a marketing one. Revenue is a vanity number. The levers that actually move income are charging enough to escape the churn cliff, plugging the leaks you already pay for, protecting the margin the tokens keep eating, and selling to buyers who stay. Build for what you keep, then go make more of it. When you have run this math on something real, the [community](https://community.okaneland.com) is where people compare the number they actually kept. ## Sources | Source | Link | |---|---| | Andreessen Horowitz, "The New Business of AI and How It’s Different From Traditional Software" (2020) | | [iconiq.com ↗](https://www.iconiq.com/growth/reports/2026-state-of-ai-bi-annual-snapshot)[bvp.com ↗](https://www.bvp.com/atlas/scaling-to-100-million)[bvp.com ↗](https://www.bvp.com/atlas/the-state-of-ai-2025)[a16z.com ↗](https://a16z.com/llmflation-llm-inference-cost/)[epoch.ai ↗](https://epoch.ai/data-insights/llm-inference-price-trends)[a16z.com ↗](https://a16z.com/questioning-margins-is-a-boring-cliche/)[chartmogul.com ↗](https://chartmogul.com/blog/good-customer-churn-rate/)[chartmogul.com ↗](https://chartmogul.com/reports/saas-retention-the-ai-churn-wave/)[stripe.com ↗](https://stripe.com/pricing)[support.stripe.com ↗](https://support.stripe.com/questions/understanding-fees-for-refunded-payments)[paddle.com ↗](https://www.paddle.com/resources/reduce-voluntary-and-involuntary-churn)[baremetrics.com ↗](https://baremetrics.com/blog/recover-failed-payments-save-lost-revenue)[taxfoundation.org ↗](https://taxfoundation.org/data/all/eu/value-added-tax-vat-rates-europe/)[scrapingfish.com ↗](https://scrapingfish.com/blog/indie-hackers-revenue)[linkedin.com ↗](https://www.linkedin.com/posts/robwalling_the-state-of-bootstrapped-saas-in-2025-is-activity-7315372913732329473-obpo)[bls.gov ↗](https://www.bls.gov/opub/ted/2024/34-7-percent-of-business-establishments-born-in-2013-were-still-operating-in-2023.htm)[stripe.com ↗](https://stripe.com/blog/stripe-atlas-startups-in-2025-year-in-review)No course-shilling, no "DM me to learn how." If a number can't be shown, we don't print it.