Andreessen Horowitz has published the seventh edition of its Top 100 Consumer AI Apps report, and this time it expands the project beyond traffic rankings with a revenue leaderboard built from consumer spending data from YipitData. In a16z’s telling, that added layer shows not just which AI products people use, but which ones are actually making money. (a16z’s report announcement; Olivia Moore’s overview)
The standout finding is how concentrated that spending appears to be. Olivia Moore of a16z wrote that fewer than 5% of Americans pay for AI, and that spending within that group is heavily concentrated. According to Moore’s breakdown, the top 1% of spenders average $903 a month on AI on consumer cards alone. In a separate summary post, a16z said that top 1% outspends the bottom 50% combined, while the median customer spends $25 a month. (Moore on spend concentration; a16z’s summary post)
Moore also said the new revenue rankings do not line up neatly with the traffic lists. She wrote that 29 of the 50 companies on the revenue leaderboard do not appear on a16z’s web or mobile traffic rankings, and that only seven companies show up on all three lists. (Moore on the revenue list split)
ChatGPT remains the clear leader
On the report’s headline competitive question, Moore said ChatGPT “remains the clear leader.” She wrote that the revenue data shows ChatGPT with a roughly 3x lead over both Claude and Gemini in U.S. paid subscribers, and said that gap is even larger globally. In a separate post, she said ChatGPT is at about 2x Gemini’s scale and 6x Claude’s scale on web, and 2.5x Gemini’s scale and 14x Claude’s scale on mobile. (Moore on U.S. paid subscribers; Moore on web and mobile scale)
The same set of posts says the race was not static through 2026. Moore wrote that Claude made gains over the previous six months, crossing DeepSeek and Grok in traffic and Gemini in subscribers. She also said Claude momentarily passed ChatGPT in monthly new U.S. paid subscribers. But in later posts, Moore said ChatGPT’s growth re-accelerated in usage and revenue, while Claude’s net new paid U.S. subscribers started declining in May and fell for roughly the next three months. (Moore on Claude’s gains and ChatGPT’s lead; Moore on slowing Claude growth; earlier Moore post on declining net new Claude subscribers)
The report’s overlap data points in the same direction. Moore wrote that most AI consumer subscribers pay for only one product, “ChatGPT,” and that only 13% of people who pay for one AI tool also pay for another. (Moore on subscription overlap)
Traffic and revenue tell different stories
A separate a16z post boiled the gap down neatly: traffic and revenue tell “two different stories.” The firm pointed to Replit as a top-10 consumer AI app by spend but a bottom-five app by traffic, a reminder that the products drawing the most money are not always the ones with the biggest casual audiences. (a16z on traffic versus revenue)
Moore’s own explanation for that split is that power users spend on products that help them automate work and build, ship, and market products. (Moore on what power users buy)
Consumer assistants may be emerging, but agents are still early
The report also points to a possible next phase for consumer AI products. Moore wrote that this edition showed signs of an emerging personal-assistants trend, with a wave of products that are now “consumer-grade,” meaning usable by non-technical people. She pointed to Muse and Instinct as early traffic leaders, even though they missed the cutoff for the main rankings. (Moore on emerging assistants)
But the same posts stop well short of calling agents mainstream. Moore wrote that mainstream agent usage is still “very early,” and said Muse’s early downloads were only a fraction of what Threads saw at launch. She also said current use of agent products is still largely centered on coding and automation workflows, though travel is starting to emerge as another category. (Moore on early agent adoption)
Reactions from people around the report underline that usability gap without changing the underlying data. a16z partner Josh Elman wrote that giving an uninitiated user an agent can still feel like handing them “a blank box,” while investor Nikita Bier argued that broad non-coder adoption now looks more like a user-experience and learning-curve problem than a capability problem. (Elman on the “blank box” problem; Bier on UX as the barrier)
The clearest takeaway from a16z’s latest report is not just that ChatGPT is ahead. It is that, by a16z’s measures, consumer AI revenue is still being driven by a relatively small group of heavy spenders, even as the category starts looking for products that can work for far more ordinary users.
