Andreessen Horowitz finds AI's top 1% of paying consumers outspend the bottom half combined

Split the spending by head and the gap comes to more than fiftyfold ^[7]. If 1% of paying cardholders outspend the bottom 50% as a group, the average heavy spender pays more than 50 times what the average bottom-half spender pays ^[7]. That floor holds before you count any skew inside either group. Put it next to the $903 August average and, assuming the ratio held that month, the typical bottom-half spender paid less than about $18 a month ^[11].

At $903 a month, the heavy cohort spends about $10,836 a year ^[8]. Eighteen months earlier the figure was roughly $502 a month ^[9]. PYMNTS wrote that these buyers spend at levels that "look less like entertainment subscriptions and more like business software budgets" ^[6]. The analysis as reported does not give the number of cardholders, dollar totals for either group, the bottom half's growth rate or any count of non-paying users.

The concentration can resolve in three ways. One possibility is that the top 1% are freelancers and small firms putting software on a personal card. In that case the revenue is business spending that happens to show up in a consumer panel, and it should renew for as long as the work pays. A second is that the cohort is simply early, and spending on building and automation tools works its way into the bottom half until the gap narrows. In the third, the cohort stays small while its bills keep rising, and revenue moves further away from any count of users.

I think the data fits the first and third readings better than the second. The 80% rise is measured inside the same top 1% ^[4], and PYMNTS described paying for AI as a much narrower habit than using it ^[5]. Getting from there to what an app is worth is an inference from concentration, because valuations are not in the data. The view would be wrong if a later cut of the same series showed the bottom 50% taking a larger share of spend. It would also be wrong if heavy spenders turned out to cancel faster than everyone else.

For app makers, the choice is where to put product money. A company that builds usage-based plans for the $903 buyer has less to spend on converting the free users who fill its headline count. At more than 50 to 1 per head, the numbers favour that bet for now ^[7].