The $903 Club: a16z's First Consumer AI Spend Ranking Reveals a Market That's Wide but Shallow

For three years, Andreessen Horowitz has ranked the world’s most-used AI products by web and mobile traffic. The seventh edition of its Top 100 Gen AI Consumer Apps report, published October 5 by partner Olivia Moore, does something different: for the first time, it ranks products by what consumers actually pay — observed spending on US consumer credit cards, drawn from YipitData’s e-receipt panel. The picture that emerges is a market that is extraordinarily wide and surprisingly shallow, with a small caste of power spenders carrying an outsized share of the entire consumer AI economy.

The headline numbers

The report’s most striking stat is a study in inequality. The top 1% of US AI payers accounted for 19.5% of all observed consumer AI spend in August 2026 — more than the bottom 50% of spenders combined, who contributed just 16.6%. That top percentile pays an average of $903 per month on consumer cards for AI products, a figure that has grown 80% over the last 18 months. The median AI payer, by contrast, spends $25 a month and has “barely expanded” that spend over time.

The broader adoption picture is similarly lopsided. Nearly half of US consumers now report using AI, but only 25% engage with it daily. As of August, just 4.5% of US consumers held an active paid personal subscription to ChatGPT, Gemini, or Claude — though that is up from 2.1% a year ago, meaning the paying base has more than doubled in twelve months. Even among payers, overlap is rare: only 13% of users who pay for one AI product pay for even one other.

Who are the $903-a-month people? Moore describes them less as mainstream consumers and more as prosumers — individuals buying software to build, create, and get work done. Across the panel, the top 1% of spenders disproportionately purchase automation and product-building tools like n8n, fal, Manus, and Nous Research’s Hermes Agent, and over-index on creative tools like Higgsfield, Figma, and HeyGen.

Three LLMs, three different games

The traffic story remains familiar: ChatGPT is still the pioneer and the leader, with roughly a 2x lead over Gemini in August web visits and a 6x lead over Claude. On mobile, ChatGPT holds a 2.5x lead over Gemini and a 14x lead over Claude in monthly active users. Per YipitData’s panel, ChatGPT has 3x more US consumer paid subscribers than Claude or Gemini combined-tier rivals.

But the more interesting shift is Claude’s emergence as the clear number three — and arguably the most aggressive monetizer. Claude did not even rank on a16z’s first web list in September 2023, yet has now passed DeepSeek and Perplexity in traffic and, earlier this year, passed Gemini on US consumer subscribers, even after Google migrated legacy paid users to its AI plan. The standout figure: 7.3% of Claude’s consumer payers are on Max, its $100/month top individual tier, versus 1.3% for Google and 1.1% for ChatGPT at their comparable $100 tiers. Anthropic, which has pledged not to run ads in Claude, has bet squarely on subscription revenue from committed prosumers.

By mid-summer, however, the winds shifted. YipitData recorded declines in Claude’s average daily sessions in July and August, with slowing gross additions and rising churn, while ChatGPT re-accelerated on the July launch of GPT-5.6 Sol, Terra, and Luna and the debut of ChatGPT Work.

What traffic rankings were hiding

The methodology change exposes how much the industry’s standard lenses miss. Of the top 50 vendors ranked by observed spend, 29 do not appear anywhere on a16z’s web or mobile traffic lists — desktop applications and agents living inside messaging platforms that Similarweb and Sensor Tower simply cannot see. Only seven companies appear on all three lists (web traffic, mobile traffic, and spend): ChatGPT, Claude, Suno, Perplexity, Photoroom, Canva, and Notion — with ChatGPT topping each one.

The edition also saw the fewest debutantes ever: just 11 first-time entrants, the smallest tally across seven lists, suggesting the consumer AI leaderboard is settling. Incumbents are now prominent — Canva and Notion rank in the web top ten, Figma in the top 15, and Google alone claims five web entries (Gemini, NotebookLM, AI Studio, Labs, and Antigravity). Notably, a16z also changed eligibility rules going forward: products primarily designed for NSFW use cases are excluded — a category that would otherwise have made up more than 20% of the web traffic list.

The personal agent land grab

The report’s third act is the race to become the consumer’s personal assistant. Multiple startup agents now report hundreds of thousands of users — Instinct, Tomo, Poke, Lindy, and Town among them — while incumbents have counterattacked with Meta’s Muse, OpenAI’s Dots, and xAI’s Grok Bot.

The numbers are getting real. Instinct, founded by ex-Sierra researcher Noah Shinn, passed 100,000 users in three weeks while still invite-only; 40% of users connect a personal credit card within three weeks, and those users spend an average of $1,300 a month through the agent, with Shinn claiming over $1 billion in annualized transaction volume already flowing through the product. Meta’s Muse, launched September 9, reportedly hit 250,000 daily actives in its first week and crossed 5 million downloads in under a month — faster than ChatGPT or Claude, though still far behind Threads’ 15 million downloads in 22 days. Platforms are choosing sides: Amazon shut off Meta’s Muse in under two weeks, while Shopify, Instacart, OpenTable, Expedia, and Ticketmaster signed official integrations.

Why it matters

The report’s deeper argument is about business models. Essentially all consumer AI revenue today comes from direct subscriptions or usage charges — 84% of top AI-native products offer subscriptions, 64% sell credits, only 14% run ads, and 2% take transaction fees. That is an inversion of the pre-AI consumer internet, where consumers were the product. High model costs made free-to-scale impossible, but the constraint is now visibly capping depth: consumers who can pay are paying a lot, while the vast majority still will not pay for software at all.

a16z’s suggested escape routes are the old ones — advertising and transactions. OpenAI said ChatGPT advertising reached a $1 billion annualized revenue run rate in August, on a base of 1.2 billion weekly active users, and personal agents like Instinct and Muse have both described plans to earn fees from transactions they originate. The next phase of consumer AI, the report implies, will be less about who has the strongest model and more about who finds a way to let the other 95% of users in the door without a paywall.

For now, the consumer AI economy runs on a narrow base of builders and creators willing to pay premium prices — the $903 club — while everyone else watches from the free tier.