On October 5, venture capital firm Andreessen Horowitz (a16z) published the seventh edition of its consumer AI rankings, adding for the first time a ranking based on what US consumers actually paid. The data reveals a market in which spending is concentrated among a small group of users. The top 1% of paying users spend an average of $903 a month, while the median across all payers is just $25. A closer look at where the money goes and how services are priced suggests that individuals who use AI heavily for work and creative projects are propping up the paid market as AI use spreads.
The "4.5% paid subscription" figure is not the share of AI users who pay
In August 2026, 4.5% of US consumers in a YipitData survey based on electronic receipts held a paid personal plan for at least one of ChatGPT, Gemini or Claude. According to a16z, that is up from 2.1% a year earlier. Paid personal subscriptions to the three major services are growing steadily, yet the vast majority of those surveyed do not pay. That is the picture presented in a16z's seventh edition.
However, it would be wrong to read the 4.5% as "the share of AI users who pay." The denominator is all consumers in the survey, and the subscriptions counted are limited to the three major services. It is not a payment rate covering every AI product, including image generation and developer tools.
Other surveys confirm that AI use itself is widespread. In a Pew Research Center survey of 5,119 US adults conducted February 17–23, 2026, 49% said they use chatbots, and 24% said they use them daily.
| Metric | Share | Population and timing |
|---|---|---|
| Use chatbots | 49% | US adults, Pew survey, February 2026 |
| Use chatbots daily | 24% | All US adults, same Pew survey |
| Hold a personal paid plan with one of the three major services | 4.5% | YipitData US panel, August 2026 |
Placed side by side, these surveys suggest that AI use is broad while personal paid subscriptions remain limited. But a survey of usage and a receipt-based check of paid subscriptions differ in respondents and timing. A conversion rate to paid plans cannot be calculated directly from 49% and 4.5%. Pew's survey serves as a point of comparison for gauging how widespread AI is, regardless of whether people pay.
Just 1% of payers outspend the bottom 50%
In consumer AI spending identified in a16z's research, the top 1% of payers accounted for 19.5% of total spending. The entire bottom 50% of payers combined accounted for 16.6%. In other words, 1% of payers pay more than the bottom 50% put together.
データを表で見る
| Share of total observed spending (%) | |
|---|---|
| Top 1% of payers by spending | 19.5 |
| Bottom 50% of payers by spending | 16.6 |
Calculating from these spending shares, the average per-person spending of the top 1% is about 59 times that of the bottom 50%. The formula is (19.5 ÷ 1) ÷ (16.6 ÷ 50), dividing each group's share of spending by its share of payers to compare per-person averages. This is a rough estimate based on figures in the seventh edition's spending analysis and does not apply directly to US consumers as a whole or to AI users worldwide.
As of August 2026, average monthly spending per person in the top 1% reached $903, up 80% over the past 18 months. Meanwhile, the median monthly spend among all payers was $25, roughly flat. The median is the middle value when spending is lined up from smallest to largest, and it is a different measure from the top 1% average. Even so, it shows that some users have sharply increased their AI spending while typical payment amounts have stayed relatively stable.
The destinations of spending are also distinctive. According to a16z, payments to the automation tool n8n, fal (which provides infrastructure for running AI models) and the AI agent Manus stand out among top spenders. In creative fields, Higgsfield, Figma, HeyGen and others are widely used. a16z sees these high spenders as individuals who also use AI for work and creative production. That means "consumer spending" paid on personal credit cards may include fees for software used for business.
In developer and creative tools, costs grow the more you use them
The pricing structure of the developer tool Cursor offers a useful reference for why AI spending can climb so high. In its official pricing, checked on October 11, 2026, individual plans are Pro at $20 a month, Pro Plus at $60 a month and Ultra at $200 a month, all before tax. If a user exhausts the usage allowance included in a plan, they can keep going with additional usage-based charges or move to a higher plan.
When an external AI model is selected, the allowance is consumed according to that model's API pricing. Even for the same number of requests, consumption varies by model. If you repeatedly hand work to an AI agent, the actual bill can't be judged from the base monthly fee alone.
Cursor's official explanation also gives rough guidelines: people who use agents daily spend $60–$100 a month, and those who frequently use multiple agents or automations spend $200 or more a month. However, these are reference figures provided by Cursor, not the results of an independent third-party spending survey.
Of course, this pricing example alone cannot explain what combination of plans the August top 1% actually held to pay $903 a month. Still, it is easy to see how individual spending can become high when usage charges are added to flat fees. Using several development and creative tools together pushes the monthly total higher still.
When people use AI for work, its price is often weighed against other software, outsourcing costs and their own working hours. For writing code, producing video and automating routine tasks, what matters is whether the results justify the money. This view is drawn from the uses and pricing structures of products people are actually buying.
However, a16z's spending data does not measure how much working time users saved or whether they earned more than they spent on AI. Paying high fees and getting results worth the cost are separate matters.
Usage rankings and spending rankings differ greatly
Of the top 50 companies by spending, 29 appeared in neither the web nor the mobile usage rankings compiled by a16z. Looking only at website visits or app users may overlook the services that actually collect the most payments.
a16z points to AI agents that run in desktop apps or inside existing messaging services as one reason for the gap. If AI can be called directly from the work apps people already use, there's no need to visit a standalone website repeatedly. And products where a small number of users run large volumes of processing and pay high fees tend to rank differently from services with many free users.
However, web and mobile usage rankings and US card-spending rankings measure different things and cover different regions. The difference in rank cannot be taken as a difference in companies' global revenue or profit. a16z itself states that the YipitData figures come from US survey respondents and do not represent companies' total revenue.
Still, examining actual payments rather than just user counts has value. Checking where, and for what purposes, a service is used and what triggers charges makes it possible to separate why many people use it from which uses generate revenue.
Another important factor is the cost on the provider's side. If high-spending users run large volumes of AI processing, the provider needs correspondingly large computing resources. As a result, higher user spending does not necessarily mean higher profit for the provider.
Ads that support free use, and trust in AI
Among the 44 AI-centered products at the top of a16z's web usage ranking that it examined, 84% used flat-rate subscriptions and 64% used usage-based billing or sold additional credits. Meanwhile, 14% of products ran ads, and 2% earned revenue from transaction or platform fees.
These are the shares of products adopting each revenue model, not shares of total revenue. Some products combine multiple revenue models.
a16z cites the high cost of running AI models as the reason charging users directly is so common. Even before usage grows enough for advertising revenue alone to sustain a business, the cost of handling free users' requests is incurred. Subscriptions and usage-based fees let providers have users bear costs from the moment service begins.
At the same time, efforts to support free use with advertising revenue have already begun to take shape as a business. On August 31, OpenAI announced that ChatGPT's ad revenue had reached an annualized $1 billion. That is a conversion assuming the current revenue pace continues for a year, and does not mean $1 billion in revenue was booked over the past year.
The company also explained its policy of clearly separating ads from AI responses and keeping ads from influencing answers. OpenAI's announcement shows that advertising is already a revenue source of some scale. However, it is not clear how much of the cost of serving free users that revenue covers.
Anthropic, by contrast, has taken a different approach. In its policy of not introducing ads, announced on February 4, it explains that Claude's revenue comes from enterprise contracts and paid subscriptions.
The company's concern is that emphasizing ad revenue could cause the best answer for users to diverge from the provider's profit. Part of the value of conversing with AI lies in accomplishing a goal quickly. But trying to extend usage time to increase ad revenue can conflict with that value. This is a design concern raised by Anthropic, and it does not prove that other companies' ads are actually distorting answers.
If ads and transaction fees spread as revenue sources, AI users alone would no longer have to bear the costs. But the cost of AI processing does not disappear, and the question arises of whose interests the AI is operating to serve.
Going forward, it will be necessary to see how long high spenders keep increasing their spending, how many people take out paid personal plans, and how much of the cost of free use can be covered by revenue such as advertising.
If value for users and profitability for providers can both be achieved, the paid AI market could expand beyond the small group of individuals who use it intensively for work to a much wider range of users.