Are new AI apps still making money?

Are people still spending more money on AI apps today?

Yes. Spending on AI apps is still climbing very fast today, so there is no sign that users have broadly stopped paying for them.

Sensor Tower’s 2026 State of AI research gives us the cleanest consumer-level evidence. Global in-app purchases for generative-AI apps reached $6.1 billion from Q2 2025 through Q1 2026, up 232% from the comparable previous period. Quarterly revenue went from less than $60 million in early 2023 to about $1.9 billion in Q1 2026. That is more than a 30-fold increase in roughly three years.

The growth kept going after that. Sensor Tower estimated more than $4 billion of AI-app in-app purchases in the first half of 2026, 36% above the previous six months. Usage grew with the spending: time spent in generative-AI apps was projected to rise from 17.2 billion hours in the first half of 2025 to 36 billion hours one year later.

That combination is much stronger than a burst of novelty downloads. People are opening these products more, spending longer inside them and paying more.

AI has become important enough to affect the entire mobile economy. In 2025, spending in non-game apps surpassed games for the first time, according to Sensor Tower, with generative AI among the main categories driving the shift.

So the first part of the question is easy to settle. There is more money going into AI apps these days, not less.

Is ChatGPT taking nearly all the money from new AI apps?

ChatGPT takes an enormous share of consumer AI spending, but plenty of money is still reaching smaller and newer AI products.

The concentration at the top is striking. Sensor Tower estimated that ChatGPT generated almost $1.3 billion of mobile revenue in Q1 2026 alone. That was more than five times the combined revenue of the other nine apps in its top-ten AI ranking.

General AI assistants are therefore a particularly tough market for a newcomer. A startup building another general chatbot has to compete with OpenAI, Google, Anthropic, xAI and several other companies that already have huge user bases, model budgets and distribution.

Outside general assistants, the market is much less settled. AI companion apps generated around $150 million in Q1 2026, more than 12 times their quarterly revenue three years earlier. Sensor Tower also found far more fragmentation in AI agents and image-and-video generation, with spending spread across a wider group of products.

Menlo Ventures found a similar pattern from a different angle. Its consumer-AI research estimated roughly $12 billion of annual consumer spending on AI tools. General assistants captured most of that spending, while specialized tools received around 19%.

Nineteen percent of a multibillion-dollar market still leaves a large business opportunity. The harder conclusion is that a specialized AI app needs a good reason to exist separately from ChatGPT, Claude or Gemini. Generic access to intelligence is already dominated by giant platforms.

Do AI apps actually make more money per customer than normal apps?

Yes. AI subscription apps currently monetize users better than non-AI apps, especially during the first year.

RevenueCat’s 2026 State of Subscription Apps report analyzed more than 115,000 subscription apps representing over $16 billion in revenue. Its AI-versus-non-AI comparison is unusually useful because it looks beyond famous startups and compares large groups of ordinary apps.

The median AI app converts 2.4% of downloads into paying users within 35 days, versus 2.0% for non-AI apps. Trial-to-paid conversion is 8.5% for AI products against 5.6% elsewhere.

The revenue difference is even clearer. Median realized first-year lifetime value per payer reaches $30.16 for AI apps, compared with $21.37 for non-AI apps. That gives AI roughly a 41% advantage.

People are therefore showing a real willingness to pay extra for AI functionality. The problem appears later, when those customers have to decide whether the product deserves to remain in their monthly budget.

Median metric	AI apps	Non-AI apps
Download-to-paid conversion	2.4%	2.0%
Trial-to-paid conversion	8.5%	5.6%
First-year revenue per payer	$30.16	$21.37
12-month monthly-plan retention	6.1%	9.5%
12-month annual-plan retention	21.1%	30.7%
Refund rate	4.2%	3.5%

How much money does the typical new app actually make?

Very little. The huge AI success stories give a badly distorted picture of what a normal new app earns.

RevenueCat measured monthly revenue one year after launch across recent subscription apps. The median was about $72 a month. Reaching $429 a month was enough to enter the top quartile, while roughly $2,574 a month put an app in the top 10%.

That means the threshold for the top decile was around 36 times the median.

The longer-term milestones look equally harsh. RevenueCat found that only about 4.6% of newly launched subscription apps reach $10,000 in monthly recurring revenue within two years. Around 1.7% reach $25,000.

These figures cover subscription apps broadly rather than AI alone, so we should not pretend they describe the exact income distribution of every new AI product. They still give us a much better baseline than startup funding announcements because AI has become deeply embedded in the subscription-app market. RevenueCat classifies 27.1% of the apps in its dataset as AI-powered, rising to 61.4% in Photo & Video and 41.1% in Productivity.

There is also a revealing age effect. Apps launched before 2020 still generate 69% of all subscription revenue in RevenueCat’s dataset. Apps launched in 2025 or later account for only 3%.

So when somebody asks whether new AI apps are making money, we need to distinguish possibility from probability. The possibility is obvious. The probability of launching an app and quickly turning it into a meaningful business remains low.

Are brand-new AI apps still reaching $100 million in revenue incredibly fast?

Yes. Even now, a small group of young AI apps is reaching nine-figure annualized revenue at speeds that would have looked absurd in previous software cycles.

Lovable is one of the clearest cases. The AI app-building company crossed $100 million in annualized revenue within months of its public launch. By early 2026 it had passed $400 million. In June, Lovable told TechCrunch that it had exceeded $500 million in annualized revenue while users were creating around one million new projects every week.

Higgsfield has moved even faster lately. The AI image-and-video startup said in August that it had reached $700 million in annualized revenue and 30 million users across 200 countries. It also said 390 Fortune 500 companies were using its products. Earlier in the year, its annualized revenue had been closer to $200 million.

Cognition provides an even fresher example. In September, the company behind Devin said its annualized run-rate revenue had risen from $492 million to about $900 million in only four months.

Three companies, three different use cases and the same unusual pattern: AI apps can still go from young startup to hundreds of millions of dollars in reported annualized revenue extraordinarily quickly.

These are extreme outliers. We should read them alongside the $72 median for ordinary new subscription apps, because together those figures reveal what has become distinctive about this market: the ceiling is enormous while the middle remains tiny.

Can we trust all these huge AI app revenue numbers?

We can trust them as evidence of strong customer demand, but we should be careful about treating every reported “ARR” figure as equivalent.

The terminology around private AI-company revenue has become loose. Traditional SaaS ARR usually annualizes recurring subscription contracts. AI companies increasingly have subscriptions, consumption revenue, enterprise commitments, credits and other payment structures running together.

Some companies report annualized revenue based on a recent month. Others describe annualized run-rate revenue. Committed ARR can include contractual spending that has yet to become recognized revenue. These numbers can all be useful, but they do not measure exactly the same thing.

Cognition gives us a good current example. Its $900 million figure is explicitly annualized run-rate revenue. TechCrunch noted that the company did not disclose exactly how it calculated the metric. Higgsfield also describes its $700 million figure as annualized revenue. Lovable reports annualized revenue run rate.

We therefore should not add $900 million from Cognition, $700 million from Higgsfield and $500 million from Lovable and pretend we have measured $2.1 billion of audited trailing revenue. What we can say is simpler: three very young products are receiving enough customer spending at their current pace to annualize into hundreds of millions of dollars.

That is already remarkable. Inflating the claim beyond what the disclosures support would only make the analysis weaker.

Has AI made apps so easy to build that making money is actually harder?

Yes. AI has made building an app dramatically easier, while getting people to notice and keep using that app has become more difficult.

RevenueCat’s launch data captures how quickly supply has changed. Around 2,000 new subscription apps were launching each month at the beginning of 2022. By early 2026, the figure had climbed above 14,700 a month.

That is roughly seven times as many launches in four years.

The timing matters. RevenueCat says the sharpest acceleration on iOS began during the rise of AI-assisted development tools. Those tools did not create the whole increase, but coding agents, vibe-coding platforms, templates and inexpensive model APIs clearly lowered the amount of engineering needed to ship a usable product.

Meanwhile, more than 200,000 apps now mention AI in their descriptions according to Sensor Tower. Calling something an “AI app” no longer gives it much novelty by itself.

This is one of the biggest changes for new founders. A few years ago, building the product could consume most of the effort. Today, somebody using Lovable, Replit, Claude Code, Cursor or similar tools can reproduce surprisingly sophisticated functionality quickly.

The scarce part has moved toward finding the right problem, getting distribution and giving users enough reason to return after the first impressive demo.

Is enterprise AI where new apps are making the most serious money now?

Yes. Enterprise AI is currently one of the clearest places where new applications can turn a narrow use case into a large business.

Menlo Ventures estimated that companies spent $37 billion on generative AI in 2025, up from $11.5 billion one year earlier. Around $19 billion went to applications rather than the underlying models and infrastructure.

That application spending is already spread across several markets. Menlo estimated $8.4 billion for horizontal AI tools, $7.3 billion for departmental applications such as coding and sales, and $3.5 billion for industry-specific products.

Startups captured roughly 63% of AI-application spending in Menlo’s model, up from 36% the previous year. That result is particularly interesting because enterprise software usually gives established vendors a huge advantage. Microsoft, Salesforce, Google, Adobe and other incumbents already have customers, procurement relationships and distribution. New AI companies are still taking a large part of the spending.

Coding has become the biggest departmental category, reaching an estimated $4 billion. Healthcare, legal work, sales, customer support, design and marketing are creating their own substantial markets.

This tells us something more useful than simply observing that companies have AI budgets. Businesses are increasingly paying independent AI applications to perform identifiable work inside existing departments.

Enterprise AI application segment	2024 spending	2025 spending	Approx. growth
Horizontal AI	$1.6B	$8.4B	5.3×
Departmental AI	$1.8B	$7.3B	4.1×
Vertical AI	$1.2B	$3.5B	2.9×
Total applications	$4.6B	$19.2B	4.2×

Which new AI apps are finding it easiest to charge customers?

AI apps tied to expensive, repeated work are finding the clearest path to serious revenue.

Coding is the strongest example so far. Menlo estimates companies spent about $4 billion on AI coding products in 2025. Lovable is monetizing software creation for people who often cannot code traditionally. Cognition sells autonomous engineering work through Devin. Replit has pushed far beyond its original collaborative coding product into AI-assisted software creation.

The pattern also shows up in creative work. Higgsfield sells AI image and video production and has increasingly pushed into marketing teams and large enterprises. Gamma reached $100 million in annual recurring revenue while remaining profitable, according to the company, by turning AI into presentations, websites and other finished communication assets.

Healthcare gives us another version of the same idea. Menlo estimated roughly $1.5 billion of spending on vertical healthcare AI, including around $600 million for ambient clinical documentation. Doctors already spend large amounts of time producing notes, so the economic value of reducing that work is easy for a hospital or medical group to understand.

AI legal products such as Harvey have followed a similar path. Instead of asking law firms to pay for generic artificial intelligence, they attach AI to research, drafting, review and other legal work that already consumes expensive professional hours.

Across these categories, customers can see what they are buying. They receive software, a finished video, a presentation, a medical note, a legal draft or completed engineering work. That makes pricing much easier than selling vague access to “AI.”

Can a simple AI wrapper still make money today?

Yes, a relatively simple AI product can still make money today, although staying simple makes the business increasingly fragile.

The word “wrapper” sometimes hides how software businesses have always worked. Many successful products rely on infrastructure they did not create. An AI app does not automatically become worthless because it calls OpenAI, Anthropic, Google or another outside model.

What matters is how much value exists around that call.

A product can still win through interface design, specialized prompts, proprietary context, integrations, distribution or a very narrow workflow. Even a technically simple tool can become valuable if it reaches the right audience and saves enough time.

The danger appears when the entire product can be copied by changing a prompt or when the same capability appears inside ChatGPT, Claude or Gemini.

Granola is a useful example of how an AI app can respond. It originally became popular as a cleaner way to take AI meeting notes. Meeting transcription has since become crowded and increasingly commoditized. Granola has been moving deeper into team workspaces, shared company knowledge, APIs and connections with other AI tools. The company is trying to own what happens to the meeting information after transcription.

That direction makes sense because RevenueCat’s retention numbers are unforgiving. AI subscriptions already lose customers faster than conventional apps. A thin product that can be replaced in five minutes has even less room for error.

Simple AI apps can still work. The ones that grow tend to become less simple as soon as they find something customers genuinely care about.

Are cheaper AI models helping new apps make more money?

Usually yes. Falling model costs give good AI apps room to do more work for the same customer price, although competitors receive the same advantage.

A few years ago, model cost could kill an otherwise useful product. Long conversations, image generation, video, large documents and multi-step agent workflows could become expensive very quickly.

Inference keeps getting cheaper while models keep getting stronger. An app can now process more context, call models repeatedly, generate several outputs and automate longer workflows for a cost that would previously have been difficult to support.

That improves the economics of products sitting well above the underlying model layer. If a customer pays $50 because an app saves several hours of professional work, cutting inference cost from $10 to $4 makes the business considerably better without forcing the company to lower its price.

The effect is much weaker when the app sells something almost identical to the model itself. Competitors get the same cheaper inference, and customers can increasingly access strong models directly.

Higgsfield shows why costs still matter, especially at the heavy end of AI. When the company announced its latest funding round, CEO Alex Mashrabov described video as one of AI’s most compute-intensive categories and said reliable compute capacity had become a major operating requirement. Massive annualized revenue can therefore coexist with massive infrastructure spending.

Revenue alone still cannot tell us whether an AI app is a great business. We also need to know how much it costs to serve heavy users and how much revenue remains after model, compute, acquisition and support costs.

Are people subscribing to AI apps and then quickly cancelling?

Yes. AI apps have a genuine retention problem right now, and it is probably the biggest warning sign in the whole dataset.

RevenueCat found lower 12-month retention for AI apps across every subscription duration it studied. Monthly AI subscriptions retain 6.1% of customers after a year, compared with 9.5% for non-AI apps. Annual AI plans retain 21.1%, compared with 30.7%.

Refunds are also higher: 4.2% for AI apps versus 3.5% elsewhere.

Those numbers become more interesting when placed next to AI’s stronger conversion and revenue-per-payer figures. People are more willing to try AI, more willing to become paying customers and more valuable during their first year. They are also more likely to leave.

That sequence fits what we can observe in the market. A new AI product can produce an extraordinary first impression. Users subscribe, solve the immediate task and later discover that they do not need it often enough, that another tool offers the same capability, or that their general AI assistant has learned to perform the job.

The problem is retention rather than lack of curiosity.

This is also why impressive launch revenue deserves less weight than it did in conventional SaaS. For an AI app, the interesting question begins after the first successful paywall.

Are OpenAI, Google and Anthropic making independent AI apps pointless?

No. Independent AI apps still have plenty of room, but competing directly with the big AI assistants has become a terrible place to start.

General assistants are heavily concentrated. ChatGPT reached one billion monthly active users on mobile in about three years, according to Sensor Tower. ChatGPT, Gemini, Claude, Grok and other major assistants also keep absorbing capabilities that once supported separate startups: research, files, coding, image creation, voice, search and increasingly agent-like work.

A new company therefore needs to think seriously about platform risk. If its entire product can become one menu option inside a major assistant, its position is weak.

The application market nevertheless shows that specialization still works. Menlo’s enterprise research estimated that AI-native startups captured 63% of application spending despite competing with some of the largest software companies in the world. Startup share reached 71% in product and engineering tools.

Large AI platforms are built for a huge range of users. A focused application can spend all of its energy on one profession, one workflow, one dataset or one type of output. It can also build collaboration, permissions, integrations and industry-specific behavior that a general assistant may have little reason to prioritize.

The opening remains very large. It is simply much narrower for products whose only advantage is access to a good model.

What do the AI apps making real money have in common?

The strongest new AI apps tend to sit directly inside a job people already value enough to pay for.

Look at where the biggest application businesses are appearing. Cognition tackles software engineering work. Lovable turns instructions into working applications. Higgsfield creates marketing and video assets. Gamma creates presentations and visual material. Harvey works inside legal workflows. Clinical AI scribes remove documentation from doctors. Granola captures and reuses information from meetings.

These products vary enormously in technical complexity, customer type and pricing. The recurring feature is easier to see from the customer’s side: each app produces something useful that previously required meaningful time, skill or labor.

Distribution then separates many of the winners from technically similar products. As building becomes cheaper, having comparable AI quality becomes less unusual. Reaching a particular professional group, fitting into an existing workflow and becoming habitual can be much harder to reproduce.

Retention completes the picture. RevenueCat shows that AI can get customers through a paywall unusually well. Keeping them is where the market breaks apart.

That gives us a useful test for a new AI product. We should be able to explain, in ordinary language, which repeated task it handles, why somebody pays for that task today, how the app reaches those people and why they will still need it six months later.

If those answers are weak, adding a stronger model usually will not rescue the business.

So, are new AI apps still making money?

Yes. New AI apps are still making money today, and the best ones are making extraordinary amounts of it.

The latest evidence is unusually consistent. Consumer spending on generative-AI apps keeps rising. AI subscription apps generate 41% more first-year revenue per payer than non-AI apps. Enterprise spending on AI applications has moved into the tens of billions. Lovable recently passed $500 million in annualized revenue, Higgsfield reported $700 million, and Cognition has just reported roughly $900 million in annualized run-rate revenue.

Anyone claiming that the window for making money from new AI apps has already closed is fighting the data.

The tougher finding is about what happens to everybody else. Monthly subscription-app launches have risen roughly sevenfold. The median new subscription app earns only around $72 a month after its first year. AI apps churn faster than conventional apps. General-purpose AI is increasingly controlled by a few giant platforms. Model capabilities that once looked like standalone products can become standard features surprisingly quickly.

Those facts can coexist because the AI app market has become extremely uneven. The average outcome is small, while the right tail is enormous.

Our conclusion is therefore quite specific: there is still a lot of money available for new AI apps, probably more than ever in absolute terms. Getting a piece of it has become harder because simply building an impressive AI product has lost much of its scarcity.

The strongest opportunities today are attached to expensive, repeated work where customers care about the result far more than the model underneath it. Coding, video production, healthcare documentation, legal work and other professional workflows already show that pattern.

A founder can still launch an AI app now and build a very large business. A founder launching “an AI app” without a strong answer to who pays, what recurring job gets done and why users stay is entering a much harsher market.