Consumer AI Adoption Stalls at 64% While Spending Triples to $40B

A market growing in depth, not breadth

New consumer AI data released this fall seems to contradict itself at first. Usage is close to universal at the top: ChatGPT alone reported 1.2 billion weekly users at OpenAI's DevDay on September 29, up from 900 million in February.24 Spending is rising quickly as well. Menlo Ventures estimates that global consumer AI spending reached $40 billion in 2026, more than three times its $12 billion estimate for the year before.11 Card data from U.S. banks, though, shows that only a small share of households pay an AI company directly.29

The numbers are less contradictory than they look. Consumer AI has stopped growing mainly by adding users. It now grows mainly by getting more money from people who already rely on it, and most of that money comes from a fairly small group of heavy users. That shift has consequences for how OpenAI and its competitors design their apps, set prices and pay for the computing power their products need.

Adoption has flattened

The clearest evidence comes from Menlo Ventures' second annual State of Consumer AI report. It is based on a survey of 5,067 U.S. adults run with Morning Consult in July 2026.11 The share of American adults who use AI went from 61% to 64%, a gain of three points.11 Menlo puts global growth in users at about 11%, from 1.8 billion to 2 billion people.13 Amy Wu Martin, a Menlo partner, said the main surprise was that the market more than tripled "without a wave of new users."15

The people already using AI are using it more. Daily use rose from 19% to 25% of U.S. adults.14 Among AI users, daily use went from 31% to 39%.13 Menlo notes that general AI assistants by themselves are now used daily by 22% of consumers. That is more than ride-sharing and food delivery combined, and among the habits Menlo tracks, only streaming is clearly more common.13

The opposite group has not moved either. The share of Americans who do not use AI dropped only three points, to 36%.14 Among those non-users, 70% now distrust what AI tells them, up from 58% a year earlier. Seventy-six percent cite privacy concerns, and 83% say they would rather deal with a person.11 Menlo's own summary is that both ends of the market became more extreme while the middle stayed about where it was.15

Why the payment numbers differ so much

Estimates of how many people pay for AI range from roughly 2% to 55%. The gap mostly comes from what each source measures, and the details matter.

Menlo reports that 55% of U.S. AI users pay for at least one AI product, which the firm equates to about 90 million Americans.13 That count includes plans paid by someone else, such as an employer. A footnote in Menlo's report says 48% of payers pay for at least one tool themselves.11 Combining those two figures gives about a quarter of AI users paying out of their own pocket. That is still a large share, but it is much smaller than the headline number.

Bank transaction data is far lower. PNC's internal card data showed 2.2% of its households paying for a generative AI subscription in May 2026, with those households spending an average of about $31 a month, up from about $22 two years earlier.9 Some outlets that relayed the PNC figure through Andreessen Horowitz's September State of Markets report dated it to April instead of May.5 A careful analysis of the PNC numbers points out that they describe PNC's own customers, not a nationally representative sample, so a headline about "2.2% of U.S. households" overstates what the data shows.9 Bank of America Institute data put paying households at about 3% in 2026, with a median monthly spend of $20. The number of paying households was 38% above the 2024 average.2

A third measure, cited in reporting on a Bloomberg analysis, found that paid personal subscriptions to major chatbots in the U.S. reached 4.5% in August 2026, up from 2.1% a year earlier.10 OpenAI's own figures fall between these. It reported more than 50 million consumer subscribers alongside 900 million weekly users in February, a conversion rate of about 5.6%.22 Measured against September's 1.2 billion weekly users, the same subscriber count would be closer to 4%, and OpenAI has not released a newer number.24

Taken together, the most reasonable conclusion is that self-reported survey payment and bank-observed direct payment are answering different questions. Neither one is wrong. Read side by side, they suggest that a large share of the paid AI people use is covered by employers, schools or other household members, and that only a small minority of households has chosen to pay for a chatbot themselves. Streaming makes a useful comparison: PNC measured it in about 43% of households, against about 2% for AI.9

A small group pays most of the bill

The most important finding for business models is how concentrated spending has become. Menlo reports that the 14% of payers spending $100 or more a month account for about 60% of U.S. consumer AI spending.14 Coverage of the Bloomberg analysis describes a similar pattern: the top 1% of AI spenders average about $903 a month, while the median payer spends around $25.10 Bank of America found that 7% of AI-paying households now spend more than $100 a month, and that the share spending $21 to $40 has grown 50% since 2024.2

The people paying most are the ones who use AI most. Menlo found that payers are almost twice as likely as non-payers to use AI every day (50% vs. 26%). They are also five times as likely to use AI agents (65% vs. 13%).11 Of the people who use agents, 92% pay for AI.14 Over the whole user base, 41% have tried an agent, 24% use one regularly, and 32% have let an AI take an action without giving final approval.11 Shawn Carolan of Menlo described those actions as minor so far, such as sending an email or making a purchase, but said they show people beginning to hand over judgment.15

This creates a structural problem for the industry, which coverage has also pointed out. Agents are expensive to run, and their users are concentrated among a small number of heavy payers. That leaves a revenue base that is narrow and uneven compared with the cost of serving it.3

How ChatGPT's tiers reflect the split

OpenAI's product changes this year map closely onto this divided market. In January it expanded its $8-a-month Go tier worldwide and said it would test ads for U.S. users on the free and Go plans.33 Ads began appearing on February 9. They are labeled as sponsored, are kept off sensitive topics such as health and politics, and are not shown to minors. Plus, Pro, Business, Enterprise and Education plans do not show ads.31 Reuters called the decision a major departure for a company that had relied on subscriptions. It noted OpenAI's plan to spend more than $1 trillion on infrastructure by 2030 and analysts' warnings that ads could push users toward Gemini or Claude.33

At the other end, prices have risen sharply. One industry tally of current plans lists consumer chatbot tiers from Google's $4.99 AI Plus to a $500-a-month ChatGPT Pro tier, and finds that 9 of 19 paid tiers cost $99.99 or more.1 Menlo describes the approach as using "the subscription to price the access and the meter to price the excess."1 Ads cover the cost of the large free audience, low-cost tiers convert casual users, and premium and metered plans collect most of the revenue from the power users who generate most spending.310

This approach depends on the free audience accepting ads. Subscriptions currently account for about 85% of revenue from leading consumer AI web products, and advertising for only 13–14%.10 Early performance claims for ChatGPT ads have come mainly from marketing-industry commentary rather than from audited disclosures, so they should be treated with caution.3435 Trust is the bigger risk. Menlo found that AI users now rank accuracy, trustworthiness, and security and privacy above ease of use when choosing a product.11 An ad-supported assistant now has to show it meets those standards.

Outlook

The direct-payment numbers do not point to a weak market, and the $40 billion figure does not mean mass monetization has arrived. Consumer AI has become a widely used, mostly free utility with a small, intensely engaged group of paying users. Paying households are spending more and staying longer: PNC found the average subscription lasted about seven consecutive months, up from about five at the start of 2024.9 The broader base has not decided to pay.

The main question for the coming year is whether the next step in growth comes from getting more households to pay, from charging heavy users more, or from advertising to everyone else. The evidence so far points to the second and third. If that continues, the AI assistant experience will increasingly depend on what users pay: ad-supported for most people and premium-priced for a minority.