95.5% of Americans won’t pay for AI. The people who do spend $903 a month.
a16z published the seventh edition of its Top 100 Gen AI Consumer Apps report on October 7, 2026, and for the first time the list is ranked by something better than downloads: US card spending. Download counts tell you what people will try for free. Card data tells you what people will pay for when the free trial runs out. Those turn out to be very different products.
The headline numbers are blunt. Only 4.5% of Americans subscribe to any major chatbot. Almost nobody pays for AI. And the ones who do, pay hard — the top 1% of AI spenders average $903 a month.
That gap is the whole story. Not “will consumers pay for AI?” but “which tiny slice pays a lot, and what are they buying?”
The $903 crowd isn’t chatting
When a16z looked at where the heavy spenders actually put their money, chat didn’t top the list. Their dollars flow into HeyGen and Higgsfield (video and motion tools) and into builder platforms like Manus and n8n.
There’s a pattern in that: every one of those tools replaces a line item somebody was already paying for. Video production, ad creative, workflow automation, an ops contractor. Nobody spending $900 a month is curious about AI. They run payroll against it.
If you build or sell AI products, this is the uncomfortable question from the data: does your product replace a budget line, or does it just make an existing free activity slightly faster? The first kind gets $900 a month. The second gets churned.
ChatGPT has the subscribers. Claude has the margin.
The rivalry numbers surprised me. ChatGPT holds three times the US paid subscribers of every rival combined — a gap that looks unassailable until you read the next finding. Claude caught up to Gemini in US consumers this summer, and it got there on the back of high-tier plans: people paying more for better work output.
Read that as a pricing lesson rather than a scoreboard. The growth didn’t come from winning the $20 trench war at the bottom. It came from people trading up — coders and analysts who found the cheaper tier was the expensive choice because bad output costs more than a subscription does. The low end of AI pricing is a bloodbath. The premium tiers are where the margin hides, and Claude’s summer is the receipts.
Subscriptions still beat ads, for now
84% of the AI-native web apps on the list sell a subscription. Only 14% run ads. That ratio isn’t ideology — it’s physics. Ad models need massive, repeatable attention, and most AI surfaces haven’t figured out attention economics yet. A chat window doesn’t hold still the way a feed does.
Worth watching, though: OpenAI launched a visual ad format inside ChatGPT image generation just this week, with a full measurement stack behind it. If anyone can make AI surfaces behave like ad inventory, it’s the company with 1.2 billion weekly users. The 14% may not stay 14% forever. But if you’re picking a business model this quarter, the card data says subscriptions still win.
Agents are next — for five million downloaders and about twelve power users
The report flags personal agents as the consumer front, with Meta’s Muse as the proof point: 5 million downloads in under a month. Then comes the asterisk. Actual usage stays concentrated in power users, the people who enjoy rewiring their week more than they enjoy the result.
That’s fine, and normal. New interfaces start as hobbyist toys. So was the App Store. I just wouldn’t build a plan around agents staying niche forever. The real planning question is what happens when a Muse-class agent ships by default on the next billion phones — at that point you find out whether agent usage is a preference or just a friction problem waiting to be removed.
What to do with this
If you price an AI product: stop optimizing the $20 tier first. Build a tier for the $903 person, price it like a work tool, and name it after the line item it replaces. The barbell says the money is at the ends, not the middle.
If you sell to businesses: your buyers are farther along than your roadmap. They already replaced costs with AI; they need someone to make those replacements visible and defensible. Audit what your team quietly automated this quarter — there is always more than the official count.
If you build content or marketing: 95.5% of your audience uses the free tier. Optimize for the moment they hit the wall — the failed render, the mangled export, the twelfth regenerate — because that’s the moment the card comes out.
If you’re watching the space: track the spend data, not the download charts, over the next two quarters. Downloads flatter consumer apps. Cards don’t.
The full report is worth your time: a16z’s Top 100 Gen AI Apps, 7th edition.
