THE MONEY IN AI ISN’T GOING TO THOSE WITH THE MOST USERS. IT’S GOING TO THOSE WHO UNDERSTAND WHAT PEOPLE WILL PAY FOR.

I’ve read the latest Andreessen Horowitz report — the seventh edition of Top 100 Consumer AI Apps. And do you know what surprised me most? Not another set of ChatGPT records. Not even Claude’s progress.

It was the sheer gap between AI’s popularity and where the money actually goes.

While some count users by the million, others quietly make money serving a few thousand customers. And perhaps it’s the latter who understand the market far better.

Let’s take a closer look.

1. Has OpenAI won? For now, yes. But not everywhere.

ChatGPT remains the undisputed leader of the mass market.

Its web traffic is twice Gemini’s and six times Claude’s. On mobile, the gap is even wider: 2.5 and 14 times, respectively.

But the most interesting developments concern Claude.

Anthropic has already overtaken DeepSeek and Perplexity in traffic, and earlier this year it passed Gemini in the number of U.S. paying subscribers.

Moreover, 7.3% of Claude’s paying users choose the expensive Max plan, starting at $100 a month. The comparable share for ChatGPT is just 1.1%.

What does this mean?

OpenAI is winning the mass market. Anthropic is aggressively competing for professionals who use AI as a working tool.

And those are two very different economic models.

Another intriguing number: only 8% of U.S. paying ChatGPT subscribers also pay for Claude.

Increasingly, people are choosing a tool for a particular way of working rather than “the smartest model.”

So I wouldn’t rush to declare an outright winner. Leading in audience size does not necessarily mean winning on profitability.

2. The report’s defining number: $903 a month

This is where the real economics of AI begin.

Only 4.5% of U.S. consumers have a paid personal subscription to ChatGPT, Gemini or Claude. Yet around 25% already use AI every day.

But let’s look at who brings in the money.

The highest-spending 1% of paying users spend an average of $903 a month!

That 1% accounts for 19.5% of observed consumer spending on AI.

The bottom half of paying users accounts for just 16.6%.

The median user spends $25 a month. And that figure is barely growing.

The picture that emerges is striking.

One person needs AI to write a greeting, summarize an article or generate an image.

Another uses it to develop software, automate a business, create advertising campaigns, design products and, in effect, replace the work of an entire team.

For the first person, $20 is already a meaningful expense.

For the second, $900 is a small amount if the tool helps them earn or save several thousand dollars.

People don’t pay for intelligence. They pay for the economic results it helps them achieve.

To me, that is the market’s real dividing line.

3. The most popular products are not always the ones bringing in the most money

Here is another number that should give startup founders pause.

Of the top 50 vendors by consumer spending, 29 do not appear at all in the rankings of the largest AI products by web traffic or mobile audience.

Let that sink in.

We are used to judging startups by traffic, downloads, MAU and registrations.

Yet the money may be somewhere else entirely.

In tools for coding, automation, video generation and professional creative work.

Suno, ElevenLabs, Cursor, OpenRouter and specialized medical services all show that you don’t have to build another general-purpose ChatGPT.

You need to solve a specific problem better than anyone else.

A thousand customers paying $500 each means $500,000 a month. You don’t need millions of users to do that.

Of course, revenue is not profit. AI products can incur substantial model and infrastructure costs.

But the direction is clear: selling measurable utility is a better business proposition than trying to appeal to everyone.

4. The next big battle is not over subscriptions. It’s over the right to spend the customer’s money.

Take personal agents.

Instinct, which works through iMessage, gained 100,000 users in three weeks.

According to its founder, 40% of users link a payment card. Those who make purchases then spend an average of $1,300 a month through the agent.

Its reported annualized transaction volume already exceeds a billion dollars.

But there is an important detail here.

A billion dollars in transaction volume is not a billion dollars in revenue.

For now, the agent helps users spend money, while the company plans to earn fees on those transactions.

Imagine how this could develop.

The agent finds and books a hotel. Buys tickets. Orders products. Chooses insurance. Arranges car repairs.

And earns a fee on every transaction.

This is no longer the chatbot market.

It is a battle for the position between consumer and seller.

Google, marketplaces, aggregators and advertising platforms used to occupy that position.

Now AI agents are trying to break into it.

And that changes the very architecture of e-commerce.

5. The AI market is maturing. And startups will have a harder time.

The new ranking has just 11 newcomers. That is the lowest number across all seven editions.

The major players are consolidating their positions. Google, OpenAI, Anthropic and Meta are actively expanding their capabilities.

The idea of “putting an API behind a nice interface and selling a subscription” is becoming less and less viable.

Any sufficiently popular feature could be built into a foundation model developer’s own product tomorrow.

What does that leave independent companies?

Things that are difficult to replicate with a simple model update:

• Unique data and accumulated expertise.
• Deep integration into real business processes.
• Control of equipment, manufacturing and logistics.
• Specialization in a particular industry.
• Customer trust and responsibility for results.

And this is where I see a very interesting opportunity.

6. Where would I look for money in AI today?

Certainly not in building yet another general-purpose chatbot.

I would go where money is already being spent, mistakes are costly and processes are inefficient.

Take car repair. An agent does more than read diagnostic trouble codes: it helps the mechanic devise a sequence of checks and pinpoint the fault.

Or manufacturing. An agent analyzes process parameters, formulations and the causes of defects, then proposes ways to optimize equipment settings.

Or legal practice. The system does more than summarize legislation: it helps prepare procedural documents and develop a strategy for a particular case.

Or small business. An agent does more than reply to a customer: it manages the deal, working with the CRM, documents, invoices and suppliers.

In all these cases, what matters is not a polished conversation with AI.

What matters is the result: how much was earned, how much was saved and how many errors were prevented.

And I would prefer to sell such solutions through a clear economic model rather than by token count: implementation costs, ongoing service, payment for completed operations or a share of verified savings.

7. And the main takeaway

Right now, many entrepreneurs are trying to make money from AI’s popularity.

I think that is the wrong way to frame the problem.

The opportunity is to make money from what AI makes possible.

ChatGPT may have an audience of a billion. Claude may attract the professional users who spend the most. Personal agents may gain control over consumer transactions.

But for a small business, none of that matters very much.

You don’t need to beat OpenAI. You need to find a thousand customers for whom your agentic system creates real economic value.

And perhaps a hundred would be enough.

One more thing. I would not turn the statistic about the 1% spending $903 a month into dogma. It is the average spending of the highest-spending U.S. consumers who pay for AI, not proof that every professional customer is willing to pay you $900.

But the signal is unmistakable.

The AI market is ceasing to be a contest between smart models. It is becoming a contest for control over real work, real decisions and real money flows.

That is where the big money will be made.

And anyone who keeps selling “access to artificial intelligence” risks discovering that their main competitor already offers a similar feature for free.

Source: Andreessen Horowitz — Top 100 Consumer AI Apps, Seventh Edition, October 5, 2026.

Note: the spending figures are based on YipitData research panels, primarily covering the United States; they do not represent total company revenue.

#ArtificialIntelligence #AI #AIAgents #Business #Automation #Startups #AIEconomics