🤖 People aren’t just using AI anymore—they’re paying for it. But not many are paying.
Andreessen Horowitz (a16z) data reveal an interesting feature of the consumer AI market: widespread use doesn’t yet mean widespread payment.
For the first time, a16z has added data on observed U.S. user spending to its Gen AI Consumer Apps ranking, based on a YipitData panel.
And the key signal here is not just the amount spent, but who exactly is paying.
FACT
In August 2026, only 4.5% of participants in the relevant US panel had an active personal paid subscription to at least one of the three services—ChatGPT, Gemini, or Claude.
A year ago, the figure was 2.1%.
In other words, the paying audience is growing, but it remains narrow for now.
The distribution of spending is even more interesting:
🔹 the top 10% of payers account for about half of all observed spending;
🔹 the top 1% account for 19.5% of spending;
🔹 the bottom 50% account for just 16.6%;
🔹 the top 1% spend about $903 per month on average;
🔹 the median payer spends about $25 per month.
Moreover, spending by the top 1% rose substantially: by about 80% over 18 months—from $504 in January 2025 to $903 in August 2026.
And median spending barely changed over this period.
This is an important point: AI economy growth is still largely being driven by the top tier of users, not the typical payer.
And $903 is not “the average American’s spending on AI,” nor is it company revenue. It’s the figure for observed spending by the top 1% of payers in a specific YipitData panel.
MECHANISM
What’s happening?
A divide is emerging between the everyday AI user and the power user.
For the first person, AI may be a tool to:
— ask a question;
— translate text;
— create an image;
— write an email;
— get advice.
For the second person, AI is gradually becoming part of their work infrastructure.
It is among the biggest spenders that a16z sees disproportionate demand for coding, productivity, and creative tools.
That’s why the monetization model is changing.
Users aren’t simply paying for “AI.”
They pay for a specific task that AI helps them complete faster or better.
This is no longer just technological curiosity.
AI is gradually becoming a work tool.
At the same time, the claim that users pay specifically to “save time” or “earn additional income” is an analytical interpretation, not a direct a16z metric.
CHATGPT ≠ THE WHOLE MARKET
ChatGPT remains the undisputed leader.
According to a16z, it has about three times as many paid consumer subscribers in the US as Claude or Gemini individually.
But the picture is different when it comes to reach.
In web traffic, ChatGPT leads Gemini by about 2× and Claude by about 6×.
On mobile, the figures are approximately 2.5× and 14×, respectively.
This shows something important:
audience, usage, and monetization are not the same thing.
29 of the 50 products ranked by observed spending did not also make the top 50 for web traffic or mobile audience.
In other words, a large audience ≠ big money.
For example:
🔹 Suno ranks 19th in web traffic, but 7th in spending;
🔹 ElevenLabs ranks 25th in web traffic and 10th in spending.
A vertical AI product may have a smaller audience, but much greater value for a particular user.
RISK
It’s easy to draw the wrong conclusion here:
“Americans spend $903 a month on AI.”
No.
$903 is the figure for the top 1% of payers in the panel studied.
YipitData is not a census of all US consumers, and observed spending cannot be equated with the total revenue of specific companies.
a16z also notes that some of this spending may be work-related. That means actual professional use of AI could be even higher.
There’s another risk for the market itself.
If most spending is concentrated among a small group of power users, that still doesn’t prove AI has become a mass-market paid product.
For now, the model mostly looks like this:
subscription + additional usage-based payments.
In a16z’s ranking, 84% of AI-native web products use subscriptions, 64% offer additional payments or credits, 14% use advertising, and just 2% rely on transaction or platform fees.
So, for now, AI is largely being sold as software.
But the next phase may be different.
DECISION
For users, the question should not be:
“Which AI is the hottest right now?”
A:
“Which AI delivers a measurable result for me?”
If a service costs $20 and genuinely saves time, its value can be quantified.
If you’re paying for 5–10 AI services at once but none delivers a clear result, subscription inflation sets in.
For businesses, the signal is even more interesting.
The winner in the AI market may not be the one with the most users.
The winner will be the one that becomes an indispensable tool for a specific job.
And this is where the next phase may take shape:
chatbot → agent → digital worker.
While the mass-market user is experimenting with AI, a small group is already making it part of their own economy.
FACT → MECHANISM → RISK → DECISION
DVA — facts without the hype.
Not financial advice.
Preserving values and time.
