Is AI Profitable? Data From 98 AI Projects
Yes, AI can be profitable, but adding a model call to an app does not make it a business. In ProvenStartups’ full cohort of 98 AI projects, 26 publish a…
Yes, AI can be profitable, but adding a model call to an app does not make it a business. In ProvenStartups’ full cohort of 98 AI projects, 26 publish a clean monthly figure; the median is $17K/mo and the range is $300/mo to $115K/mo. ProvenStartups would build a narrow workflow or productized service with visible buyer value, and refuse to build a generic wrapper whose only advantage is its prompt.
Contents
What 98 AI projects actually show
The useful answer is not in funding announcements. It is in the full matching set of 98 projects, including 80 solo-run businesses. Of those, 26 disclose clean monthly results, with a $17K/mo median. That median describes the cohort, not just the named examples below, and it is not a forecast for a new launch.
The cohort contains 48 AI services, 25 AI content businesses, 15 AI websites, and 10 AI e-commerce projects. This matters because “AI business” is not a single model. A consulting engagement, a content pipeline, and a self-serve app have different sales motion, support load, and margins.
The revenue ceiling is real but easy to misread. nano-banana.ai reported approximately $115K/mo in net profit for one month [C], while StoryShort.ai reported $35K/mo across three apps [F]. One is creator-relayed and one is founder-reported; neither should be presented as third-party verification.
The cohort metadata records 4 [V], 0 [F], 0 [C], and 0 [U] entries in its evidence split. That does not classify all 98 rows, so it cannot support calling the entire cohort verified. ProvenStartups keeps that caveat visible through its grading method.
Where AI businesses make money
AI revenue appears in products, retainers, projects, and commerce, but the evidence quality varies sharply. The strongest starting point for a solo builder is a painful, repeated workflow with an identifiable payer. ProvenStartups would prioritize a service that can be sold before automation, then convert repeated delivery steps into software only after demand is proven.
| Case | Model | Published economics | Grade | Difficulty |
|---|---|---|---|---|
| nano-banana.ai | AI website | ≈$115K/mo net profit for one month | [C] | 1/5 |
| StoryShort.ai | AI website portfolio | $35K/mo across three apps | [F] | 3/5 |
| AI Solo E-commerce | Information-arbitrage dropshipping | Claimed $180K in 30 days | [U] | 3/5 |
| AI digest sites | Sponsorship, membership, or traffic funnel | Author says it is not a money-making project | [U] | 2/5 |
| QuizzerAI / StudySnap | AI flashcards | No verified revenue disclosed | [U] | 2/5 |
| Claude Code SEO service | Products plus client retainers | $5,000+ cumulative from products; retainers of several thousand per month | [F] | 3/5 |
| AI voice agent | Per-client retainer | $500–$1,500/mo per client | [C] | 3/5 |
| AI lead generation | Per lead or monthly | No fixed range disclosed | [C] | 3/5 |
| AI content repurposing | Monthly retainer | No specific amount disclosed | [C] | 2/5 |
The table supports a clear decision: sell an outcome, not “AI.” A receptionist that answers calls or a service that produces qualified leads has a budget owner. A demo may prove that code works, but the StudySnap case shows that a working demo with no verified revenue [U] is not proof of a market.
Where the data contradicts the popular claim
The popular claim is that AI profit mainly comes from launching an effortless wrapper and collecting passive SaaS revenue. ProvenStartups’ cohort contradicts that story: AI Service is the largest category with 48 projects, versus 15 AI Websites. That does not prove services earn more, but it does show that the observed field is service-heavy, not wrapper-dominated.
The contradiction gets sharper when implementation difficulty is included. Across 266 software/SaaS products in the full 406-project index, difficulty counts are 12 at 1/5, 100 at 2/5, 104 at 3/5, 40 at 4/5, and 10 at 5/5. Most are not filed as effortless builds.
Services also expose pricing before a founder invests in product infrastructure. AI consulting and audits use a framework of roughly $5,000 per audit and about $50,000 per implementation [C]. Those are creator-relayed reference points, not verified average deal sizes, but they are more concrete than claims that AI profitability is “unlimited.”
The full index includes 38 cautionary tales alongside the wins. Any answer to “is AI profitable” that removes documented failures is marketing, not analysis.
What are the best AI apps to build?
The best AI apps remove a measurable bottleneck for a specific buyer: missed calls, expensive content production, slow research, or repetitive client delivery. ProvenStartups would choose a narrow workflow with human-review fallbacks and an existing budget. It would reject a broad “assistant for everyone,” an undifferentiated content generator, or a marketplace without proven supply and demand.
The site records tool mentions across 211 distinct projects. ChatGPT appears in 100 cases, Claude Code in 50, Cursor in 46, Bolt in 40, no-code tools in 23, Lovable in 19, Bubble in 15, Replit in 12, n8n in 10, and Copilot in 7. These counts show implementation choices, not which tool causes profit.
For a developer deciding what are the best AI apps, the practical order is:
- 1.Productize a service with a buyer and repeatable deliverable.
- 2.Automate the most expensive delivery step.
- 3.Add self-service only when onboarding and quality control are stable.
- 4.Keep model providers replaceable where switching costs are low.
Pricing power should come from the outcome. The AI copywriting case benchmarks traditional agencies at $10K–$15K/mo [C], but that reference does not establish what a new AI copy app will earn.
How to test AI profitability before building
Test the sale before the stack. Get a buyer to agree on the deliverable, price, and acceptance criteria; then measure inference, review, support, acquisition, and failure costs. ProvenStartups would not extrapolate from a viral launch month, treat gross revenue as profit, or use an unverified creator claim as a base-case forecast.
Use this sequence:
- 1.Pick one buyer and one expensive recurring task.
- 2.Deliver it manually with AI assistance.
- 3.Track every paid input against OpenAI’s published API pricing.
- 4.Record human review and exception handling, not just token cost.
- 5.Check adoption context against the U.S. Census Bureau’s report on business AI adoption.
- 6.Automate only the steps that repeat across paying customers.
Avoid models whose attractive economics exist only on a slide. The AI-to-AI marketplace proposes a 10%–30% match commission [U], but the take rate does not prove liquidity, transactions, or revenue. It is also rated 5/5 difficulty.
Site-wide, 106 cases publish a clean monthly figure: 8 are under $1K/mo, 18 are $1K–$10K/mo, 54 are $10K–$100K/mo, and 26 exceed $100K/mo. That is a distribution of disclosed cases, not startup odds. Build the forecast from signed customers and measured costs.
FAQ
AI is profitable in some documented cases, but the range is too wide and the evidence too mixed for a universal promise. The right questions are whether a specific buyer pays, whether delivery remains reliable, and whether the cited revenue is verified, founder-reported, creator-relayed, or unverified.
Is AI profitable for a solo founder?
Yes, it can be. In the 98-project AI cohort, 80 projects are solo-run; across the full index, 246 of 406 are operated solo. The strongest cited upside is nano-banana.ai at approximately $115K/mo net profit for one month [C], but its creator-relayed grade and single-month scope make it a case, not an expected result.
How much money can an AI app make?
Among all 98 matching projects, 26 publish a clean monthly figure. Their median is $17K/mo, ranging from $300/mo to $115K/mo. Those figures answer what disclosed projects have reported, not what a new app will make. StoryShort.ai’s $35K/mo across three apps [F] is useful evidence, with founder-reporting risk still attached.
Are AI services or AI SaaS more profitable?
The dataset does not disclose a clean category-level profitability comparison, so claiming a winner would invent evidence. It does show 48 AI Service projects versus 15 AI Websites in the cohort. A service can validate demand faster; the AI voice-agent framework prices at $500–$1,500/mo per client [C], but that is creator-relayed.
What AI business should a developer start?
Start with a narrow, paid workflow that can be delivered manually before it is automated. ProvenStartups would favor lead generation, voice reception, local-business SEO, or content repurposing over a generic chatbot. The Claude Code SEO case reports $5,000+ cumulative product revenue plus retainers of several thousand per month [F].
How should AI revenue claims be verified?
Read the evidence grade beside the number. Across ProvenStartups, 57 cases are third-party verified [V], 184 founder-reported [F], 121 creator-relayed [C], and 44 unverified [U]. Preserve the original scope too: revenue is not profit, a single month is not recurring history, and “no revenue disclosed” must never be converted into an estimate.