The Two Prices of a Minute
The internet turned our attention into money. Artificial intelligence is preparing to sell us our time back. That difference may determine who wins in the new economy.
A minute has two different prices.
The first is what an advertiser pays to capture our attention during that minute. The second is what we are willing to pay to get something done without spending it.
Much of the internet economy grew around the first price. For social media, video platforms and digital advertising, time spent on screen became a valuable commercial asset.
Artificial intelligence could change that calculation.
Completing a four-hour financial analysis in half an hour, shortening a software development task that would otherwise take hours, or investigating a subject far faster than the days of research it would normally require...
Can we measure the value of these things solely by the minutes spent in an app?
The seventh edition of the Top 100 Gen AI Consumer Apps report, published on 5 October 2026 by leading venture capital firm Andreessen Horowitz (a16z), offers findings that raise precisely this question.
Most used or most paid for?
For the first time, a16z compared AI applications not only by web traffic and mobile user numbers, but also by US consumer spending.
An interesting picture emerged.
Of the top 50 companies ranked by spending, 29 appeared in neither of the top 50 lists based on web traffic or mobile usage.
In the same study's US consumer card data, the top 1% of spenders accounted for 19.5% of total spending. The bottom 50% accounted for just 16.6%.
More strikingly, the top 1% spent an average of $903 a month. The median user's spending was $25.
It is hard to explain that gap through enthusiasm for technology alone.
Automation, software development and creative production tools feature prominently among these high spenders. People are buying AI not only to obtain information, but also to carry out their own work and projects.
There is a substantial economic difference between getting an answer from an application and using it to complete work that generates income.
The comparison between ChatGPT and Claude also supports this distinction.
ChatGPT leads both the usage and consumer spending rankings in a16z's research. Yet 7.3% of Claude's paying individual subscribers choose its higher-tier plan, starting at $100 a month, compared with 1.1% for ChatGPT's comparable higher-priced tier.
This does not mean Claude earns more revenue. It shows that audience size and willingness to pay more per user are different things.
So why don't customers stay?
This is where things become particularly interesting.
RevenueCat's 2026 research, covering more than 115,000 apps, finds that subscription apps with AI features generate 41% more first-year revenue per paying customer than other apps.
But their customer retention is weaker. For annual subscriptions, 12-month retention is 21.1% for AI apps, compared with 30.7% for other apps.
People may be willing to pay more at the outset, but that does not mean they will keep subscribing.
One possible reason is that an application can deliver an impressive first experience without becoming an indispensable part of daily life.
An AI tool bought to prepare a presentation may go unopened for months once the presentation is finished. A tool that analyses cash flow every month, monitors financial risks or handles a recurring operation, by contrast, can become a service people need regularly.
A tool used only once can, of course, still be valuable. The question is which revenue model can sustain that value.
The challenge for AI companies is not simply to help customers, but to turn that benefit into a service worth buying again.
What users gain is not the same as what companies earn
Research published by Stanford University in April 2026 illustrates the scale of this difference from another angle.
By measuring how much US users would require in compensation to give up access to generative AI tools, the researchers estimate annual consumer benefits of approximately $172 billion.
This figure is neither AI companies' revenue nor a direct measurement of time saved. It is an estimate of the economic value people place on the technology.
That is what makes it interesting.
A technology can deliver enormous economic benefits to people. How much of that benefit becomes company revenue, and how much becomes company profit, is a separate question.
Of the 44 AI-focused web products examined by a16z, 84% offer subscriptions, while 64% offer usage-based pricing or additional credits. Only 14% use advertising. Companies can combine these models.
Heavy AI usage also carries a computing cost. A user who pays hundreds of dollars a month is therefore not necessarily a correspondingly profitable customer.
From a financial perspective, the question is clear: how much of the value created can be captured as revenue, and what does it cost to produce that value?
High growth figures alone are therefore not enough. Customer loyalty, the cost of serving customers and sustainable profitability matter just as much.
The next stage: delegating decisions, not just saving time
AI no longer merely answers our questions. It researches transactions, compares options and increasingly acts on our behalf.
According to Menlo Ventures' 2026 survey of 5,067 US adults, 32% of AI users have allowed an AI tool to complete a transaction on their behalf without their final approval at least once.
This development also opens the door to a new revenue model: charging for completed transactions.
But it brings another issue, just as important as the technology, into focus: trust.
If AI selects the most suitable financing offer on your behalf, for example, what will it prioritise? Your total financing cost, or the commission it earns from the transaction?
As we begin to delegate not only our time but also our decision-making authority to AI, conflicts of interest, auditability and accountability will become much more important.
When evaluating the successful AI companies of the future, we may therefore need to add other measures alongside visitor numbers, time spent and subscription revenue: the value of the work completed, the accuracy of the results, whether customers trust the same system with the task again, and the true cost of the service.
Today, heavy AI users often spend more time with these tools. There is not yet enough evidence to say that screen time is generally declining.
But as the technology completes more work independently on the user's behalf, time spent in an application may become less useful as a standalone measure of success.
Perhaps some of the most valuable applications of the future will be those we trust to work in the background, rather than those we open repeatedly throughout the day.
For years, we put a price on the minutes people spent in front of screens. Now we need to learn how to value giving those minutes back.
Comments