Consumer AI is a power-user game: 4.5% pay, 10% fund most

1. Spend reshuffles the list — ▶ 1:33

   Traffic is settling: only 11 products were new across the web and mobile lists, the fewest in seven editions. Spend tells a different story, since 29 of the 50 products ranked on spend were on neither traffic list. About half of Americans use AI, but around 4.5% pay, double a year ago. The top 10% of payers deliver over half of revenue, the top 1% about 20%, the bottom 50% about 16%. The top 1% spends $93 a month, the median payer $25. Heavy spenders skew to n8n, Granola, Higgsfield and Manus.

2. Agents: OpenClaw fades, successors surge — ▶ 3:31

   OpenClaw would have ranked high last report but is now absent as traffic collapsed, and its team was acquired by OpenAI. Grockbot, Town and OpenAI's Dots have overtaken it, and Moore says these are the first she'd hand to a family member. Instinct reported 100,000 users growing 10% a day, 40% connecting a credit card in three weeks, and over $1,000 spent in month one. Muse had 500,000 downloads and 250,000 active users in 12 days, but about 5 million US and Canada downloads against Threads' 16 million over 22 days.

3. Trust blocks network effects — ▶ 8:44

   No assistant has unlocked person-to-person network effects. The more an agent knows about you, the more useful it is, and the less you want it in contexts with other people. Many users already keep separate work and personal ChatGPT accounts, and "no one has quite cracked that yet." Elman says losing trust over email, life and credit card access "is going to be the biggest impingement" on mass adoption. Platform access cuts both ways: Amazon won't let Muse buy on its site, while Muse has hundreds of partnerships including Shopify.

4. Coders explain agent costs — ▶ 11:43

   Founders posting serving costs of hundreds or thousands of dollars per user per month looked baffling. David Pollon's Assistant Benchmark tracks 170+ agents and a community of 1,500+ early adopters, and the number one use case is coding and technical automation, even on consumer products like Muse and Instinct. Agents aimed at mainstream consumers see costs in the tens of dollars. Elman adds that newcomers face a blank box, and because use is so personal, shared ideas that spread socially haven't yet flourished.

5. Spend time, don't save it — ▶ 18:24

   Moore relays a line from Wabby CEO Eugenia: most people aren't looking to save time, they're looking for ways to spend it. That's why social media, Netflix, TikTok and YouTube are the most-used consumer products. Most consumer AI so far does things "a little bit faster, a little bit easier," which isn't a compelling daily value proposition for most people. Elman counters that a new wave of non-coders can now make videos and products they never could, and reads the top spenders as harbingers.

6. Subscriptions are an unnatural inversion — ▶ 20:21

   Moore doesn't want the 4.5% paying directly to expand. 85% of the web list monetizes by subscription, 62% via credits or token top-ups, and only 13% via ads or other models. That is "a pretty unnatural inversion" of consumer internet history. Among the top 20 consumer subscriptions, ChatGPT is the lone AI product after 3.5 years; the rest are media or frequent-purchase platforms like Amazon and Uber. High COGS ends the old density-first playbook. Elman bets falling inference costs let ads and transaction fees work.

7. OpenAI ads reach $1B — ▶ 23:51

   OpenAI hit a $1 billion annualized ad run rate, a figure from August that Moore expects is now higher, a milestone that normally takes years. It has 1.2 billion weekly users, login with ChatGPT, and a wallet she expects, so targeting could beat Meta. Elman says commercial-intent queries make ads useful answers if clearly labeled. OpenEvidence shows the niche version, with 50-60% of US physicians. Burger's question about ad margins and unit economics goes unanswered.

8. Claude passes Gemini in subscribers — ▶ 29:50

   ChatGPT leads Claude by about 6x and Gemini by about 2x on web traffic, and has roughly 3x more US paid consumer subscribers than either. But in a US spender panel, Claude has passed Gemini in paid subscribers despite Gemini's bigger base and Google distribution. Moore credits launches like Claude Design and press from the Department of War episode in February. Anthropic's no-ads stance means heavier subscription tiers: 7.5% of its subscribers are on the $100-plus plan, against about 1% for ChatGPT and Gemini.

9. Labs leave creative gaps — ▶ 31:44

   Labs have skipped audio, so Suno and ElevenLabs rank near the top of traffic and spend; Moore doubts labs want Suno's IP headaches. OpenAI's image model and Google's Nano Banana and Veo have taken mainstream traffic from standalone image generators. Midjourney fell off the traffic rankings but is back on revenue, since power users want a model with taste. Video is a "wild frontier" where Chinese companies that can train on any data have an advantage. Creative tools that agents can call will see tailwinds.

10. Startups beat the shipbuilders — ▶ 37:39

    Consumer products like Replit, ElevenLabs and Gamma are becoming enterprise companies fast through product-led growth; Canva took 6+ years. Incumbents struggle to cannibalize their interfaces: Google hasn't reinvented Docs, Gmail or calendar, and the labs do poorly outside their own chat and coding apps. Town's playbooks of how Moore writes create lock-in, since 99.9% sounds-like-me versus 85% is the difference between 10 seconds and 10-plus minutes of fixing. Elman rejects "wrapper": these products are "much more than just a shim on the model."

11. White space is in network categories — ▶ 45:35

    Most consumers still use AI as a search replacement, for essays and emails. Startup wins cluster around getting things done: productivity, photo and video editing, search. Wide open are network categories: dating (nobody on the list), recruiting, and social AI, which is mostly AI content posted on existing platforms. Shopping, home buying and retail may live in agents or standalone apps. Gaming and entertainment await better models, with AI micro dramas the exception. Elman says shopping needs visuals, not just conversation. Moore: spend-time products are often the biggest companies.

Consumer AI is a power-user market, and it stays one until it finds a business model that does not depend on subscriptions. About half of Americans use AI. Around 4.5% pay for it, double the share a year ago. Inside that group the top 10% generate more than half of revenue, the top 1% about 20%, and the bottom 50% about 16%. The top 1% spends $93 a month on personal cards. The median payer spends $25.

The people paying are builders. The spend rankings in the seventh a16z report surfaced products that traffic never showed: 29 of the 50 ranked on spend appeared on neither traffic list. The over-represented names were n8n, Granola, Higgsfield and Manus. Developer, productivity and creative tools dominate because their users are asking how to get leverage over their time. Traffic, meanwhile, has settled. Only 11 products were new across web and mobile, the fewest in the history of the series. The money has moved to the people who make things.

The mainstream wants something else. Moore relays a line from Eugenia, the CEO of Wabby: most people are not looking to save time, they are looking for ways to spend it. That is why social media, Netflix, TikTok and YouTube are the most used consumer products. Almost everything in consumer AI so far does a task a little faster or a little easier. That is not a daily habit for most people. The wide-open categories are the spend-time ones. Dating has nobody on the list. Recruiting is similar. Social AI is mostly people posting AI content on existing platforms. Shopping, home buying, gaming and entertainment wait on better models, with AI micro dramas the lone breakout. Many of these are network categories, and they need multiplayer products.

Subscriptions cannot reach that audience. Moore says she does not want the 4.5% to expand, which she admits is odd for a consumer AI maximalist. 85% of the web list monetizes through subscriptions. 62% use credits or token top-ups. Only 13% use ads or any model where the user is the product. That is "a pretty unnatural inversion" of consumer internet history, where the giants earn from ads and transaction fees. Among the top 20 consumer subscriptions, ChatGPT is the lone AI product after 3.5 years. The rest are media companies or frequent-purchase platforms like Amazon and Uber. The old playbook was to build density at near-zero cost and monetize later. High cost of goods ends that. Josh Elman adds that companies even fear growing too fast without meters on usage. His bet is that inference costs fall far enough for ads and transaction fees to work.

Ads are the first real test. OpenAI is at a $1 billion annualized ad run rate, a figure from August that Moore expects is now higher. That normally takes years. It has 1.2 billion weekly users, login with ChatGPT, and a wallet she expects to come. She suggests its targeting could beat Meta, because the advertiser is reaching someone in a live conversation. Elman says commercial-intent queries are where an ad is a useful answer, provided it is clearly labeled. OpenEvidence shows the narrow version, with 50-60% of US physicians on the product. Anthropic took the opposite route. It has said no ads, and 7.5% of its subscribers are on the $100-plus plan, against about 1% for ChatGPT and Gemini. Claude has passed Gemini in paid subscribers in a US spender panel, even though ChatGPT has roughly 3x more paid US subscribers than either.

Personal agents are the attempt to move beyond makers. OpenClaw sparked the wave, then fell out of the rankings as traffic collapsed and its team joined OpenAI. Grockbot, Town and Dots have followed, and Moore would hand them to a family member. Instinct reported 100,000 users growing 10% a day, with 40% connecting a credit card in three weeks and over $1,000 spent in the first month. Muse reached 500,000 downloads and 250,000 active users in 12 days. In the US and Canada it had about 5 million downloads over 22 days, against 16 million for Threads. Part of that is deliberate, because serving these users is expensive. And the cost line reveals who is using them. Founders post serving costs of hundreds or thousands of dollars per user per month. Assistant Benchmark data shows the number one use case is coding and technical automation. Agents aimed at the mainstream see costs in the tens of dollars. The gap between tinkerers and everyone else shows up on the invoice.

The obvious objection is trust, and nobody on the call has an answer. The more an agent knows, the more useful it is and the less you want it near other people. Many users keep separate work and personal ChatGPT accounts. "No one has quite cracked that yet," Moore says. Elman calls losing trust over email and credit card access "the biggest impingement" on mass adoption. Ads raise the same problem. A weight-loss pitch dropped into an intimate conversation does not build trust. Elman thinks OpenAI's clearly labeled approach avoids it. Burger asks what margins OpenAI earns on ads, and nobody has a figure. Platform access is contested too. Amazon will not let Muse buy on its site, while Muse has hundreds of partnerships, including Shopify. No assistant has yet unlocked person-to-person network effects.

The counter-case is that the value sits in the software layer, not the model. Elman rejects the word wrapper. These products are "much more than just a shim on the model." Town builds playbooks of how Moore writes. The gap between an email that is 99.9% her voice and one that is 85% is 10 seconds of fixing versus 10-plus minutes. That is lock-in. Big companies struggle to compete here. Google has not rebuilt Docs, Gmail or calendar, and the labs do poorly outside their own chat and coding interfaces. They leave audio to Suno and ElevenLabs. Midjourney fell off the traffic rankings but returned on revenue, because power users want a model with taste.

What is left open is whether anyone builds the spend-time products. Moore says they are often among the biggest companies. So far the money has gone to tools that save time, for the few who pay. Whether the products that reach everyone else arrive before inference costs fall far enough to fund them is not settled.