AI’s Top 1% of Payers Spend $903 a Month, a16z Data Shows

Andreessen Horowitz has been ranking the biggest consumer AI apps for three years. Its seventh edition adds something more useful than another traffic leaderboard: observed spending on U.S. consumer cards.

That changes the picture.

Nearly half of U.S. consumers now report using AI, but only 25% use it daily. In YipitData’s U.S. e-receipt panel, just 4.5% of eligible consumers had an active paid personal subscription to at least one of ChatGPT, Gemini, or Claude in August 2026, up from 2.1% a year earlier.

Among AI payers, the median spend was $25 a month.

Then there is the top 1%.

According to a16z’s seventh Top 100 Gen AI Consumer Apps report, the top 1% of AI payers spent an average of $903 a month on consumer cards. They accounted for 19.5% of all observed consumer AI spending, more than the bottom 50% of spenders combined at 16.6%.

Their spending also rose 80% over the previous 18 months.

Apparently “AI subscription” can now become a line item between rent and a car payment.

The report’s more useful conclusion is that consumer AI has found its first real market among people who pay for tools that help them build, create, and get work done. Reaching everybody else may require different products and different ways to make money.

The top 1% pays more than the bottom half

The concentration is hard to miss.

a16z, using YipitData’s U.S. e-receipt panel, found:

• 4.5% of eligible consumers had a paid ChatGPT, Gemini, or Claude subscription in August, up from 2.1% a year earlier.
• Only 13% of users who pay for one AI product also pay for even one other AI tool.
• The median AI payer spent about $25 a month.
• The top 1% spent an average of $903 a month.
• That top 1% accounted for 19.5% of observed consumer AI spending.
• The top 10% accounted for roughly half of observed spending.
• The top 1% increased spending by 80% over 18 months, while median spending barely moved.

That is a very different market from the one suggested by app-store rankings or website traffic.

a16z describes these high spenders as prosumers: people buying software to build, create, and get work done. They are not necessarily representative of the average AI user.

That distinction matters. The report’s own data says broad AI usage has arrived much faster than broad willingness to pay.

The biggest spenders are buying tools to build things

The spending data gives us a clue about why some users are willing to pay so much.

The top 1% disproportionately bought automation and product-building tools such as n8n, fal, Manus, and Nous Research’s Hermes Agent. They also over-indexed toward creative products including Higgsfield, Figma, and HeyGen.

That makes the $903 figure easier to understand. These users are often paying for tools that help them produce work.

This is also why traffic rankings increasingly miss important AI businesses.

Twenty-nine of the top 50 vendors by observed spending did not appear on either a16z’s web or mobile traffic rankings. Only seven companies appeared on all three lists: ChatGPT, Claude, Suno, Perplexity, Photoroom, Canva, and Notion.

We saw an earlier version of this divergence when we covered a16z’s sixth consumer AI ranking. Usage tells you who is showing up. Spending tells you which products have become valuable enough that users keep paying.

Those are increasingly different questions.

Claude is a good example of traffic vs. intensity

ChatGPT still dominates raw scale.

a16z says ChatGPT had roughly twice Gemini’s web visits, six times Claude’s web visits, and 14 times Claude’s monthly active mobile users in August. YipitData’s panel also showed ChatGPT with about three times as many U.S. paid consumer subscribers as Claude or Gemini.

Claude’s paying users, however, skew toward more expensive plans.

About 7.3% of Claude’s consumer payers are on its Max plan, which starts at $100 a month. Google’s corresponding $100 tier captured 1.3% of its payers, while ChatGPT’s captured 1.1%.

That fits Anthropic’s recent product focus. a16z says Anthropic has concentrated heavily on prosumer products such as Claude Design, Code Review, and Claude Science.

Claude also passed Gemini in U.S. consumer subscribers earlier in 2026. By July and August, though, YipitData’s global desktop panel showed average daily sessions declining, while its U.S. e-receipt panel showed slower gross additions and rising churn.

So Claude demonstrates both sides of the report’s thesis: a product can attract users who pay aggressively without automatically winning the broadest consumer market.

Subscriptions solved one problem and may create another

Consumer AI companies had a practical reason to start with subscriptions and usage charges.

AI models are expensive to run. a16z argues that earlier consumer internet companies could often subsidize free usage long enough to build scale because users were comparatively cheap to serve. High model costs have made that harder for AI companies.

Among the 44 AI-native products in a16z’s web ranking:

• 84% offered subscriptions.
• 64% used usage charges or extra credits.
• 14% used advertising.
• Just 2% earned revenue through transaction or platform fees.

Companies can use more than one model, so those percentages overlap.

This works especially well when users can justify the cost through their work. It becomes a tougher path to mass adoption when most people still cannot or will not pay directly for software.

a16z expects the economics to get easier as cheaper and open models improve. Product builders can increasingly route routine tasks to cheaper models and reserve expensive frontier systems for the jobs that actually need them.

That is one reason stronger open-weight models matter: lower model costs give consumer products more room to reduce usage costs and experiment with business models beyond another monthly subscription.

Personal agents could change the business model

a16z points to personal agents as one possible path.

An assistant that actually books travel, orders food, or makes purchases can potentially earn money from transactions instead of charging the user another seat fee.

The early numbers are notable, but they are company-reported.

Instinct founder Noah Shinn told a16z that 40% of users connect a personal credit card within three weeks. Among users who make a purchase, average spending reaches $1,300 a month through Instinct.

Shinn also claims more than $1B in annualized transaction volume already flows through Instinct, with about half tied to travel.

If that behavior holds at scale, the economics of a personal agent could look more like a marketplace or travel service than a normal software subscription.

Neither Instinct nor Meta’s Muse is taking transaction fees yet, according to the report. Both have described plans to do so over time.

That is the shift to watch: whether consumer AI can move from “pay us every month for intelligence” toward earning money when the software actually helps complete a transaction.

One important caveat: this is a panel, not the whole economy

The $903 number is memorable, but it needs its label attached.

YipitData’s spending figures are U.S.-only and come from consumer panels, not a census. a16z explicitly warns that the data should not be read as total company revenue.

The report also notes that work usage may be even higher than the consumer-card spending it observes.

That matters because the heaviest spenders look like prosumers. Some purchases captured on personal cards may still support professional work.

So the report is best read as a view into how a panel of U.S. consumers pays for AI, not a complete accounting of the AI economy.

The next consumer AI winner may need a different revenue model

The first wave of consumer AI monetization has been straightforward: charge directly for access, more usage, or better models.

That has produced a market where AI usage is broad, but paid adoption remains much narrower. A relatively small group of power users accounts for an outsized share of observed spending.

a16z expects subscriptions to remain a natural fit for many AI products. It also argues that advertising and transaction fees could help products reach people who will not pay another monthly fee.

The signal to watch is whether that mix actually changes.

If the share of paying consumers broadens, or if ads and transaction revenue become meaningful, consumer AI starts looking less like premium software for prosumers and more like a mass-market internet business.