Reporting this week put ChatGPT close to a billion weekly users, about seven months behind the timeline OpenAI originally set for that milestone.
Read that twice. The fastest-adopted consumer software product in modern history missed its own forecast by the better part of a year.
Nothing about that is a failure. Approaching a billion weekly users is an extraordinary outcome by any standard that existed three years ago, and it sits alongside ChatGPT taking nine percent of search. But the gap between the plan and the result is the most useful piece of information in the AI market right now.
Early velocity is not a growth rate
The mistake sitting underneath that missed target is one I have watched companies make for sixteen years, in affiliate, in ecommerce, in SaaS, and now here.
A product launches into pent-up demand and grows explosively for a few quarters. Somebody draws a line through those quarters, extends it forward, and calls it a projection.
What that line actually captured was the enthusiast population being absorbed. Those users were waiting for the product to exist. They required no persuasion, no onboarding budget, and no change to their habits.
Once that group is consumed, the next cohort is structurally different. They are indifferent rather than eager, they need a reason, and acquiring them costs real money. The curve does not break at that point. It bends, which is far easier to miss.
Where this shows up in your own numbers
Almost every AI business plan I have reviewed in the past eighteen months contains the same unstated assumption: adoption keeps compounding because the technology keeps improving.
Those two things are not connected. Capability and adoption move on separate clocks, because adoption is gated by habit, procurement, training, and trust, none of which respond to a better model.
If you sell AI-enabled software, the practical version of this is a repricing of your assumptions. Your enthusiast segment is probably already converted. Your pipeline from here is composed of people who have watched a demo, felt nothing in particular, and gone back to work.
That cohort converts on operational proof, not on capability claims. They want to know what it costs to run, who maintains it, and what happens when it is wrong. Fail that and you get pilots that die in the middle.
The bend also shows up in retention before it shows up in acquisition, which is why so many teams miss it. Enthusiasts forgive a rough edge because they wanted the product to exist. Mainstream users churn over the same edge without filing a complaint.
So a flat retention chart in the enthusiast period tells you very little about the next period. The behavior underneath it changed while the number stayed still.
Pricing carries the same trap. Early adopters accept usage-based pricing because they understand what drives the usage. Indifferent buyers read a variable bill as risk, and risk is the thing that stalls a deal at procurement.
Plan for the bend
Three adjustments worth making this quarter.
Separate your cohorts in reporting. Blended growth hides the bend for two or three quarters, which is exactly long enough to over-hire against it. Enthusiast adoption and mainstream adoption are different businesses with different economics, and they should not share a chart.
Move your forecast off capability milestones. A roadmap that assumes the next model release opens a new segment is a roadmap with no acquisition plan in it. Model quality has not been the binding constraint on adoption for a while now.
Budget for the unglamorous middle. The cost of converting an indifferent buyer is documentation, onboarding, integration support, and proof, and none of that appears in a plan built on viral compounding.
Then set your hiring against demonstrated conversion rather than projected demand. Headcount added on the strength of a forecast is the most expensive way to discover the forecast was wrong, because you find out two quarters late and the correction is a conversation with people.
There is a version of this market where everyone keeps extrapolating from 2024 and gets quietly surprised for six consecutive quarters. I think that is the base case for a lot of companies.
The signal is the miss, not the milestone
A billion weekly users is a remarkable number and it will be reported as one. The more valuable fact is that the company closest to this technology, with the strongest distribution and the most data about its own users, still got the timing wrong.
If they got it wrong with those advantages, your forecast is not conservative. It is optimistic and you have not noticed yet.
None of this argues against building in AI. It argues against building a plan whose only failure mode is that adoption arrives slower than the deck assumed.
Growth curves bend. Plan for the bend and you keep your options. Plan for the line and the bend makes the decision for you.