a16z's Latest Report: 29 Highest-Spending AI Products That Never Made the Traffic Charts

📍 On October 5, 2026, a16z published the 7th edition of Top 100 Gen AI Consumer Apps. I’ve followed it since the first edition, but this one made me stop and think for a long time.

Because for the first time, it answers — with real American credit card bills — the question every product builder should ask but rarely does:

Are “people use it” and “people pay for it” the same thing?

a16z’s answer is blunt: no. They are two completely different businesses.

That sentence sent a chill down my spine — because I once built a product with zero revenue. My thinking was simple: get the loop working. Data collection, rankings, login, payments — all wired up with free services, just to see if a paid loop could run. Market research? Where would users come from? Never thought about it. My creed back then was one line: build it well, and traffic will come.

Then I learned: users don’t just show up. You post on social media, and if it doesn’t resonate — or you say it in the wrong place at the wrong time — you get nothing. Books say your first 10 users require manual “sales,” phone calls even. But I didn’t even have their emails. Who was I supposed to call?

This report basically translated my failure into data.

📊 Counterintuitive #1: 29 top-grossing AI products are nowhere on the traffic charts

This edition, a16z did something it had never done before. Beyond the Web traffic chart (Similarweb) and the Mobile chart (Sensor Tower), it added a “consumer spending” chart — built on YipitData’s real US consumer credit card spending. Not who gets opened the most, but who actually gets paid.

The result is striking: of the top 50 by spending, 29 don’t appear on any traffic chart.

In other words, there’s a whole cohort of AI products with unimpressive website visits and modest app MAU — yet consumers happily swipe their cards for them. Meanwhile, some traffic darlings are quietly absent from the “who makes money” list.

Only 7 made all three lists: ChatGPT, Claude, Suno, Perplexity, Photoroom, Canva, and Notion. ChatGPT is the only #1 across all three.

So here’s the first takeaway: traffic is vanity, revenue is sanity — and the overlap between these two lists is shockingly small. Next time you look at a ranking, don’t just ask “who has the most users.” Ask “whose users actually pay.”

💳 Counterintuitive #2: The traffic king is worst at extracting big spend; the niche players are better at it

If #1 was about who makes money, #2 is about who’s better at getting people to open their wallets — and the answer may defy your intuition.

First, the traffic: ChatGPT’s web visits are 2x Gemini’s and 6x Claude’s; on mobile it’s even more lopsided — 2.5x Gemini, 14x Claude. Paid subscribers? 3x Claude or Gemini. The traffic crown is well deserved.

But when it comes to “getting people to spend big,” the script flips: 7.3% of Claude’s paying users are on the $100/month Max plan; for ChatGPT it’s 1.1%, Google 1.3%. In other words, Claude’s user base is far smaller, but its users are 6x more “generous” than ChatGPT’s.

Why? They’re fighting completely different wars. ChatGPT wants maximum scale — everyone, everywhere. Claude is betting on high-value workflows: coding, research. The data confirms the split: only 8% of ChatGPT subscribers also subscribe to Claude — people pick one primary assistant, and the ones who pick Claude pay more.

This is the sharpest footnote to “traffic ≠ revenue”: the traffic king wins on “how many people,” the niche wins on “how much they’ll spend.” Chase traffic, and you may end up with an army of free-tier users.

📈 Counterintuitive #3: The top 1% spend $903/month — more than the bottom 50% combined

As of August 2026, only 4.5% of US consumers had a paid subscription to ChatGPT, Gemini, or Claude. That’s up from 2.1% a year ago — more than doubled — but the absolute number is still tiny: nearly half of Americans use AI, yet only one in twenty pays for it.

And within that 4.5% (all US panel data below), the money is brutally concentrated:

① The top 1% of payers spend an average of $903/month on AI, contributing 19.5% of total spend — more than the bottom 50% of payers combined (16.6%) ② The median payer spends just $25/month ③ That top 1% grew their spending 80% over the past 18 months; the median user barely grew at all

a16z gave this cohort a name: prosumers. They’re not buying “chat” — they’re buying automation tools like n8n, Manus, and fal, and creative tools like Suno, ElevenLabs, and HeyGen. Put simply: they’re not buying AI, they’re buying “get the job done for me.”

One more number worth chewing on: of people who already pay for one AI product, only 13% buy a second one. It’s not that people won’t spend — they’ll only spend on what actually works, and once they pick one, they don’t switch.

So the second takeaway: the money in consumer AI isn’t with the masses — it’s with the minority who use AI as a production tool. If you want the masses’ money, first answer “why would they pay?” — and right now, most people don’t have an answer.

🔄 Counterintuitive #4: Subscriptions are hitting the ceiling; the next wave is “let AI spend for you”

That’s the present. Now for a16z’s take on the future — the part of the report that made me most restless.

a16z counted 44 AI-native products on the Web chart: 84% charge via subscription, 64% via usage fees, only 14% have ads, 2% take transaction cuts.

Compare that to the last internet era: in 2025, ads were 97.6% of Meta’s revenue and 73.2% of Google’s. Meanwhile, nearly every AI company today charges a “head tax” — $20/month, pay or don’t play.

Why? Because models are expensive, and the user base isn’t big enough yet for “burn cash now, monetize later.” But subscriptions have a ceiling: when only 4.5% of people pay, you can’t build a mass market on card swipes alone.

The shift has already begun. OpenAI’s ads business hit $1B in annualized revenue in August 2026; personal agents are turning “subscriptions” into “takes” — Instinct’s founder says 40% of users link a credit card within three weeks, and linked users spend an average of $1,300/month through it, pushing annualized transaction volume past $1B.

And platforms are already picking sides: Amazon cut off Meta’s personal assistant Muse within two weeks, while Shopify, Instacart, OpenTable, and Expedia proactively signed official agent partnerships. In the transaction era, the moat shifts from “how strong the model is” to “who owns the checkout.”

The logic is clean: old AI was “you ask, it answers” — a usage fee. Future agents are “it buys for you” — a transaction fee. From selling seats to taking a cut of transactions, this is the biggest lane change in consumer AI yet.

📍 So do small companies still have a shot?

By now, founders might ask: giants eat traffic and own the checkout — where can small companies still win? Chapter 4 of the report answers exactly that, distilling four moats that still work:

① Proprietary models: Suno ranks only #19 on Web traffic but #7 by spending; ElevenLabs is #25 by traffic, #10 by spending. You can’t out-chatbox the giants, but in vertical tasks like music and voice, “taste” itself is a moat.

② Multi-model experience: Cursor and OpenRouter don’t marry any single model — they pick the best one for the user. The smaller the gap between models, the more valuable “pick the right model + assemble the workflow” becomes.

③ Vertical audiences: OpenEvidence (medical literature for doctors) debuted at #47; privacy-first Venice at #43. The more concentrated the data, privacy needs, and workflows, the higher the switching cost.

④ New interfaces + hardware: Plaud, selling recording hardware plus subscription, debuted at #16 on the spending chart. When everyone’s fighting over the chatbox, a new entry point is a new continent.

The common thread: none of them compete with giants on “who has more users” — they compete on “whose users will pay more.” Right in line with the report’s throughline.

📝 Closing

Read it all, and the report really says one thing: AI competition has moved from “who has the most users” to “whose users pay” — and next, to “who takes a cut of the transactions.”

Back to my zero-revenue product. I can finally explain its failure in one sentence: I started by asking “can the loop run?” instead of “where do users come from?” If I did it again, I wouldn’t write code first. I’d answer three questions first: who are the users? Where do I acquire them? Is this a real need? Find new keywords, copy a proven product and run ads, go to specific communities, YouTube, small creators — figure out acquisition first, then build.

Traffic is the amplifier, payment is the validator — without the latter, the former is just a number.

So if you’re building an AI product, don’t start with “how many people will use it.” Start with: “who will happily swipe their card every month for a specific job to be done?” If you can’t answer that, don’t rush to write code.

If you never found users — and can’t find them — that product was probably doomed from the start.

“People use it” and “people pay for it” are two different businesses.