Loading page…
The new card-spending comparison gives OpenAI a broad geographic lead, not a measure of total revenue or all consumer preferences.
Listen to this story
YipitData’s card-based comparison of June–August U.S. consumer payments found OpenAI ahead of Anthropic in 47 states, while Anthropic led in California, Massachusetts, Montana and Washington, DC. The result maps spending within a two-company panel—not total revenue or overall chatbot preference—so free users and other AI providers are absent. Separately, YipitData’s e-receipt panel put active paid personal subscriptions to ChatGPT, Gemini or Claude at 4.5% of eligible consumers in August 2026, up from 2.1% a year earlier, showing paid adoption growing from a still-small base.
OpenAI captured at least 70% of the two-company spending split in eight states; Delaware led at 82%, followed by Mississippi at 77% and Alaska at 74%.
State wins are counted equally regardless of population, so Anthropic’s California win does not carry more weight than a win in a smaller state.
Andreessen Horowitz added observed U.S. consumer spending to the seventh edition of its report, which also ranks apps by web and mobile traffic.
Consumers spent more on OpenAI than Anthropic in 47 states during June through August, according to newly published YipitData findings covered by Business Insider. Anthropic led in three states and Washington, DC. The result gives OpenAI a broad geographic advantage among observed payments, but leaves out the people using either company’s chatbots for free.
YipitData collected the comparison’s spending data from consumer credit and debit cards between June and August. Andreessen Horowitz first published it in the seventh edition of its Top 100 Gen AI Consumer Apps report on October 5. This edition adds observed U.S. consumer spending to a report that also ranks products by web and mobile traffic.
The spending data comes from panels: groups whose transactions are observed, rather than a census of every consumer. Andreessen Horowitz explicitly cautions that its U.S.-only spending figures should not be read as total company revenue. The map therefore describes payments in the dataset, not the full financial contest between the two companies.
There is another boundary: this geographic comparison covers OpenAI and Anthropic, not other consumer AI providers. Free ChatGPT and Claude users do not register as spending. A preference expressed through unpaid use is invisible here, making the map a narrower measure than overall consumer preference.
Anthropic’s three winning states were California, Massachusetts and Montana; Washington, DC also favored Anthropic. California is the largest state by population, a useful reminder that counting state victories gives each state one place in the tally despite large differences in population.
OpenAI’s advantage was particularly pronounced in eight states, where it captured at least 70% of spending in the two-company comparison. Delaware had the highest reported share among those eight. The figures below describe the split between OpenAI and Anthropic, not OpenAI’s share of all consumer AI spending.
The same report provides context for how many consumers pay. Its subscription measure uses YipitData’s U.S. e-receipt panel and counts eligible consumers with an active paid personal subscription to at least one of ChatGPT, Gemini or Claude. It includes Gemini, unlike the two-company geographic spending comparison.
That share more than doubled over the year, but remained below one in twenty eligible consumers. This is a measure of paid personal subscriptions to those three services, not of everyone using AI. The geographic findings show where observed dollars favored each company; the subscription figures describe the size of a paying audience.
Story updates
The biggest consumer AI spenders are buying at a different scale from everyone else. Andreessen Horowitz’s October 5, 2026 analysis found that the top 1% of payers averaged $903 a month in August and supplied nearly a fifth of observed spending. Its seventh consumer AI report adds a spending ranking alongside web and mobile usage.
The addition changes what the rankings can reveal. Twenty-nine of the 50 largest AI vendors by observed consumer spending appear on neither the web nor mobile usage lists. A product can attract substantial payments from a narrower audience without becoming one of the most-used AI destinations.
YipitData supplies the new measure, which tracks observed payments on U.S. consumer cards. These are panel figures, not total company revenue. That boundary matters when comparing vendors: the spending list measures transactions captured in the dataset, while the other lists measure web traffic and mobile audiences.
Paid personal subscriptions to the three major assistants remain uncommon. In August, 4.5% of consumers in YipitData’s research had an active subscription to ChatGPT, Gemini or Claude, up from 2.1% a year earlier. That is more than a doubling, but it describes subscriptions to those three services—not payment for every AI product.
Most paying consumers also stick with a single product. According to a16z, only 13% of consumers paying for one AI product also paid for another. The median AI payer spent $25 a month, a figure that has changed little over time. The high-spending end therefore looks very different from the typical customer.
The top 1% accounted for 19.5% of observed consumer AI spending, exceeding the bottom half’s combined 16.6%. Their average monthly spending rose about 80% over 18 months, according to a16z’s analysis. Growth at that end contrasts with the largely unchanged median; higher spending is not a uniform shift across the paying audience.
The heaviest spenders disproportionately buy automation and product-building tools such as n8n, fal and Manus. They also favor creative products including Figma, Higgsfield and HeyGen. These choices underpin a16z’s interpretation that professional and creative work drives the high-spending segment, rather than treating those customers as simply more enthusiastic chatbot subscribers.
Only seven products appear across all three rankings: ChatGPT, Claude, Suno, Perplexity, Photoroom, Canva and Notion. The limited overlap makes the new list more than another popularity table. It identifies vendors collecting consumer payments that an audience-first view would leave outside the leading group.
ChatGPT still leads on reach. In August, it drew roughly twice Gemini’s web visits and six times Claude’s. Its mobile audience was 2.5 times Gemini’s and 14 times Claude’s. ChatGPT also had about three times as many U.S. consumer paid subscribers as either rival, so scale remains a substantial advantage.
Claude’s paying audience has a different mix. About 7.3% of its consumer payers subscribed to Max, its most expensive individual tier, starting at $100 a month. The comparable premium-tier shares were 1.3% for Google and 1.1% for ChatGPT. Those percentages describe each provider’s paying audience; they do not establish who has the most premium subscribers in absolute numbers.
Claude passed Gemini in U.S. paid subscribers in March 2026, a16z found. After Google changed its pricing plans in June, the two were neck and neck. That subscriber comparison and the premium-tier split answer different questions: one tracks the size of the paying base, while the other shows which plans those customers choose.
Direct payment remains central to the business model. Of the 44 AI-native products in a16z’s web ranking, 84% offered subscriptions and 64% charged for usage or extra credits. Advertising and transaction fees were uncommon. These categories can overlap: a service may sell a subscription and charge separately for additional use.
the-decoder.comGetting more people to pay for AI may be the wrong goal, Andreessen Horowitz partner Olivia Moore argues. In an October 9 TechCrunch interview, she made the case for free, ad-supported access rather than expecting everyone to subscribe. Her diagnosis: much of today’s paid “consumer AI” serves people buying tools to get work done.
Moore’s October 5 Top 100 Gen AI Consumer Apps report supplies the backdrop. For the first time, it adds observed U.S. consumer spending from YipitData alongside web and mobile traffic rankings. The spending figures come from panels, not a census, and should not be read as total company revenue.
In August, 4.5% of eligible consumers in YipitData’s U.S. e-receipt panel had a paid personal subscription to ChatGPT, Gemini or Claude, up from 2.1% a year earlier. But payments were concentrated: the top 10% of consumer AI spenders accounted for roughly half of observed spending.
The highest spenders make that divide sharper. The report puts average monthly spending for the top 1% at $903 on consumer cards, compared with $25 for the median AI payer. Those high spenders disproportionately purchase automation, product-building and creative tools. In the interview, Moore grouped power-user purchases into three work-oriented categories:
Product building: tools such as Lovable, Replit and Fal, which Moore places in the category of apps people pay for to build products.
Product marketing: AI ad generators such as Higgsfield and HeyGen, another category Moore identifies as attracting spending from power users.
Work management: apps such as Manus, Fireflies AI and Granola, which Moore says consumers initially buy but do not resemble traditional consumer products.
That distinction also changes how quickly companies seek business customers. Moore says Gamma, ElevenLabs and Cursor began with consumers and became majority-enterprise businesses within 18 months. She contrasts that with Canva, which she says took six or seven years to add team or enterprise plans.
Almost all of what we think of as consumer AI, I would argue, is prosumer AI.
Moore does not frame the solution as persuading everyone else to become a subscriber. She argues that Silicon Valley’s preference for paying directly reflects high incomes and access to corporate cards. Her expectation is that most people would prefer free access with ads, then decide whether to pay to remove them. That is a proposed route to adoption, not a demonstrated consumer preference.
Loading discussion...
Her report also identifies transaction revenue as an alternative to subscriptions and usage charges. It suggests personal agents could earn money through transactions rather than charging for each user’s access. The report presents that as a potential way to broaden the paying base, not an established replacement for subscriptions.
Lower costs are the other part of Moore’s argument. She says many consumer tasks do not require frontier intelligence—the capabilities of the most advanced models. Cheaper models could be enough, and she says founders are increasingly discussing building on open-source models. Her expectation is that this approach becomes more common when the model itself is not the product.
There is a catch in the current customer mix. Moore says users driving revenue are often coding or doing technical automation, tasks for which they probably do need the strongest models. The lower-cost opportunity she describes therefore depends partly on attracting people with different needs, rather than simply changing the models existing power users receive.
Moore points to social apps, dating, marketplaces, retail, travel, finance and health as categories without entrants in her top 100 list. She says those are areas she wants to see develop over the next six months. That is a statement about representation in her rankings, not proof that AI products do not exist in those fields.
The list measures the top 50 web products by monthly visits and the top 50 mobile apps by monthly active users. Generative AI must be central to the experience. Those rules define the opportunity Moore is describing: products that reach substantial consumer audiences, not every service that adds an AI feature.
Join the conversation
Explain which signal you trust more, and why.
Be the first to share a perspective or an experience.
Reader comments
Newest comments first. Replies stay oldest first.