Only 4.5% of Americans Pay for AI…

1. Key Themes

Theme: Consumer AI is a power-user market, and revenue concentration is extreme

The top 1% of payers spend ~36x the median

"The top 1% user is spending $903 per month personally on their personal credit cards on AI. And the median is spending $25." (Olivia Moore, a16z)

Only a sliver of users fund the market: "About 50% of Americans say they use AI, and about 4.5% pay for it." Within payers, "the top 10% of payers generate over 50% of the revenue, the top 1% generate 20%, and the bottom 50% generate 16%." Moore's summary: "the top 10% of users are driving this market."

Heavy spenders are builders, not casual users

"The heavy spenders build things. Olivia says developer, productivity and creative tools dominated, with n8n, Granola, Higgsfield and Manus overrepresented among top spenders."

Theme: Traffic rankings and revenue rankings tell different stories

29 of the top 50 revenue earners never appeared on a traffic list

"Of those 50 that ranked for spend, 29 of them were not on either traffic list." (Olivia Moore, a16z)

Traffic has also stagnated: "Traffic, meanwhile, has stopped churning, with only 11 new products across web and mobile combined." The caveat: revenue here is "YipitData's consumer card spend, so enterprise and SMB money stays out."

Midjourney shows power users and casual users diverge

"Midjourney fell off the traffic rankings. It still ranks #11 on revenue."

Casual image traffic went to OpenAI and Google, while power users want the model "that they feel has taste, has an aesthetic sense."

Replit shows the same gap

"Replit sits at #48 on web traffic and #6 on revenue."

Theme: Cost-to-serve and the subscription-vs-ads business model inversion

AI is monetizing through subscriptions, the opposite of most consumer giants

"Only like 13% had ads or other options where like you are the product instead of paying for the product." (Olivia Moore, a16z)

"About 85% of the web products a16z ranked charge subscriptions, another 62% sell credits or extra usage." Moore calls this "a pretty unnatural inversion." Her concern is that most people "don't have the funds or don't want to spend their funds on software."

Falling inference costs are the proposed path to ad/transaction models

"Josh says the way out runs through cost: inference prices are early in their decline, so ads and transaction fees start to work as they fall."

Coding workloads drive the $1,000+/month cost-to-serve for "consumer" assistants

"The number one use case was coding and technical automation, even for these consumer assistant products like Muse and Instinct." (Olivia Moore, a16z)

Coding and technical automation led with 103 mentions vs. 66 for travel. Moore reads "the $1,000-plus bills as the price of coding workloads." Josh adds that cost reshapes growth: "companies get nervous about growing too fast, because a product with no meters on usage can let spend run out of control."

Theme: Frontier lab competition is shifting, with willingness-to-pay beating distribution

Claude passed Gemini in US paid subscribers

"Claude has actually passed Gemini in terms of number of paid subscribers, which is kind of crazy given Gemini is a much bigger install base and user base overall." (Olivia Moore, a16z)

Anthropic's mix skews premium: "about 7.5% of its subscribers sit on the $100-plus Max plan, against roughly 1% at ChatGPT and Gemini." Caveat: this "comes from a panel of US spenders, so treat it as a US finding," and "ChatGPT still leads everything... with around 3x the US paid consumer subscribers of either."

OpenAI's ad business is scaling unusually fast

"That would previously take a company years and years and years to get to even after launching an ads product." (Olivia Moore, a16z)

OpenAI has "$1B in annualized ad revenue, from a product that has rolled ads out slowly" (an August number), with 1.2 billion weekly active users. Josh flags the risk: "an intimate chat that suddenly pitches a weight-loss product, which burns trust on contact."

Theme: Value is migrating to the software/context layer ("wrappers" are real products)

Personalization creates lock-in

"The difference between an email that is 99.9% sounds like me and even an email that's 85% sounds like me is the difference between spending like 10 seconds to fix it and like 10 plus minutes." (Olivia Moore, a16z)

Moore's Town agent "built playbooks of how she sounds that she cannot easily carry to whatever launches next month."

Pushback on the "wrapper" dismissal

"Josh's thesis is that value is moving back to the software layer, where context and community become the asset."

These teams ship rich products where the model is one component, "much more than just a shim on the model."

2. Contrarian Perspectives

An AI investor doesn't want paying users to grow

Moore invests in consumer AI yet argues the current subscription-dominated model is "backwards." Evidence: "Nearly every giant consumer company earns most of its money from ads or transaction fees," while "the top 20 consumer subscription products worldwide are mostly media companies and platforms like Amazon and Uber." Only 13% of ranked AI products use ads, and most people "don't have the funds or don't want to spend their funds on software." The implication is that a subscription-only AI market is capped.

"Consumer" AI assistants are really developer tools

Products pitched to mainstream users are being used mostly for engineering work. Assistant Benchmark data (3,101 messages, Aug 31–Sep 28, 2026) shows coding and technical automation first at 103 mentions, ahead of travel at 66. Costs for mainstream-targeted assistants run "in the tens of dollars a month," so the $1,000+ bills reflect coding workloads rather than consumer behavior. Caveat from the article: the panel is "keen tinkerers, not your neighbor."

Most people want to spend time, not save it

"Most people aren't looking to save time. They're looking for ways to spend their time." (Eugenia of Wabi, relayed by Olivia Moore, paraphrased from memory)

Evidence: "9 of 15 consumer categories have zero AI products in the Top 100" (streaming, social, dating, gaming, travel, retail, finance, real estate, jobs). Social media, Netflix, TikTok and YouTube lead usage "because people want to spend time," while most AI helps "finish a task a bit faster."

3. Companies Identified

Anthropic / Claude

Description: Frontier AI lab, maker of Claude.
Why mentioned: Passed Gemini in US paid subscribers; #2 on revenue; premium mix skews to $100+ Max plan.
Quotes: "Claude has actually passed Gemini in terms of number of paid subscribers..." / "Claude ranks #3 on web and #12 on mobile, and Anthropic ranks #2 on revenue."

OpenAI / ChatGPT

Description: Leading consumer AI platform.
Why mentioned: Dominant in traffic; $1B annualized ad business; 1.2B weekly actives.
Quotes: "That would previously take a company years and years and years to get to even after launching an ads product." / "ChatGPT still leads everything, at about 6x Claude and 2x Gemini on web traffic."

Google / Gemini

Description: Google's AI assistant and image/video models.
Why mentioned: Larger install base but fewer paid subscribers than Claude; absent from revenue ranking.
Quotes: "Gemini is absent because a16z could not separate its revenue from Google's." / "Google has not reinvented Docs, Gmail or Calendar for the AI era."

Midjourney

Description: AI image generation.
Why mentioned: Cleanest example of revenue/traffic divergence.
Quotes: "Midjourney fell off the traffic rankings. It still ranks #11 on revenue."

Suno

Description: AI music generation.
Why mentioned: #7 on revenue; labs have left audio open.
Quotes: "Does it make sense at this point for them to deal with all of the IP headaches that Suno has had to go through and launch a competing music model? I don't know."

ElevenLabs

Description: AI voice/audio.
Why mentioned: #10 on revenue vs. #25 on web traffic.
Quotes: "Suno ranks #7 on revenue and ElevenLabs ranks #10."

Replit

Description: AI coding platform.
Why mentioned: #6 revenue vs. #48 traffic.
Quotes: "Replit sits at #48 on web traffic and #6 on revenue."

Lovable

Description: AI app-building startup.
Why mentioned: Top-10 revenue product.
Quotes: "how Lovable hit $400M ARR in 14 months."

Perplexity

Description: AI search startup.
Why mentioned: Top-10 revenue product.
Quotes: "Perplexity, the $20B startup that bid $34.5B for Chrome."

Magnific

Description: AI image upscaling/enhancement tool.
Why mentioned: Proof a non-ad, non-VC consumer AI business can scale.
Quotes: "Magnific reached $230M ARR with zero ads and zero VC."

Instinct

Description: New consumer AI assistant.
Why mentioned: Strong early monetization.
Quotes: "Instinct announced 100,000 users growing 10% day over day, and 40% of users connected a credit card within 3 weeks." / "Instinct's users spent over $1,000 on average in their first month."

Muse

Description: New consumer AI assistant (US/Canada only).
Why mentioned: Strong tech-community launch but mainstream adoption lags Threads.
Quotes: "In the first 22 days or so, Threads had had about 16 million downloads and Muse is still sitting around 5 million over that same time period."

OpenClaw

Description: Early agent-wave product.
Why mentioned: Started the agent wave, then dropped out of rankings.
Quotes: "The agent wave started with OpenClaw, which Olivia says has dropped out of the rankings as its traffic fell; she thinks OpenAI acquired the team."

Town

Description: Agent wired into Slack, email and Google Drive.
Why mentioned: Moore's example of personalization lock-in.
Quotes: "Olivia uses Town... to write her emails and help with this report."

Granola, n8n, Higgsfield, Manus

Description: Productivity, automation, creative and agent tools.
Why mentioned: Overrepresented among top spenders.
Quotes: "n8n, Granola, Higgsfield and Manus overrepresented among top spenders."

YipitData

Description: Alternative-data provider.
Why mentioned: Source of the consumer card-spend panel.
Quotes: "YipitData's panel shows only 4.5% of US consumers paying, double a year ago."

Assistant Benchmark

Description: Tracker of AI agents run by David Pawlan.
Why mentioned: Source of task-usage data.
Quotes: "tracks over 170 agents and hosts group chats for 1,500-plus early adopters."

Wabi

Description: Company run by Eugenia.
Why mentioned: Source of the "spend time, not save time" insight.
Quotes: "Most people aren't looking to save time. They're looking for ways to spend their time."

Grok Imagine

Description: xAI's video generator.
Why mentioned: Cited as best video AI by the newsletter.
Quotes: "Grok Imagine as the best video AI is a case we made separately."

4. People Identified

Olivia Moore

Description: a16z consumer AI investor, author of the Top 100 Consumer AI Apps.
Why mentioned: Primary source of the analysis.
Quotes: "the top 10% of users are driving this market." / "a pretty unnatural inversion."

Josh Elman

Description: Investor on the panel.
Why mentioned: Argued inference costs unlock ads, and that value is returning to the software layer.
Quotes: "Josh calls the rollout deft, with ads carrying a plain label and reading like part of the conversation." / An assistant handed to a newcomer "feels like a blank box."

Elena Burger

Description: Host of the 49-minute conversation.
Why mentioned: Raised that agent-reachable creative software should see tailwinds.
Quotes: "models editing creative work go tool-calling across the internet, so creative software that agents can reach should see tailwinds."

Eugenia (Wabi)

Description: Founder of Wabi.
Why mentioned: Source of the time-spending thesis.
Quotes: "Most people aren't looking to save time. They're looking for ways to spend their time."

David Pawlan

Description: Runs Assistant Benchmark.
Why mentioned: Supplied task-frequency data.
Quotes: "he pulled the most-discussed tasks from 3,101 messages posted between August 31 and September 28, 2026."

Sam Altman

Description: OpenAI CEO.
Why mentioned: Cited on collapsing intelligence costs.
Quotes: "Sam Altman says intelligence got 100x cheaper in 2 years."

Ruben Dominguez

Description: Author of the newsletter post.
Why mentioned: Summarized the conversation.
Quotes: "I watched the full conversation so you don't have to."

5. Operating Insights

Build a premium tier for the top 1%

Price plans for heavy users, not just the median. Evidence: "Build a tier for the maker spending $900 a month, since the top 1% of payers carry 20% of revenue." Anthropic's "7.5% of subscribers on $100-plus plans, against about 1% elsewhere, shows what an upper tier looks like."

Meter heavy usage and plan the ad/transaction layer early

"Meter your heaviest users, and expose an API an agent can call. Coding and automation led task discussion, which points to where heavy cost concentrates."

Also: "Write down the ad or transaction layer you'll add once inference gets cheap," and "OpenAI reached a $1B ad run rate with a slow, labeled rollout, and trust decides whether yours works."

Win on context and personalization to create lock-in

"Own the context. A product that sounds 99.9% like you is hard to leave."

Also "Package the tinkerers": "Today's messy agent setups become tomorrow's products, and 9 of 15 categories still have none." Products emerge when "tinkerers at the edge work out a capability and post it on X, and 6 months later it is a startup, or 12 competing ones."

6. Overlooked Insights

Incumbents guard their interfaces, creating room for outside agents

"Olivia adds that incumbents guard their interfaces: Google has not reinvented Docs, Gmail or Calendar for the AI era, Anthropic's Microsoft 365 connector works inside the incumbent's own tools."

This suggests the AI-native workflow layer for core productivity suites remains unclaimed by the incumbents themselves.

Audio was left open by the labs, while video tilts toward Chinese companies

"Audio is the modality the labs left open, and both companies rank higher on money than on visits." / "Olivia calls video a wild frontier tilted toward Chinese companies, which can train on any data."

Modality-by-modality lab behavior (images contested, audio abandoned, video IP-advantaged abroad) is a useful map for where startups can outrun the labs.