AI is thought to be the biggest disruption ever in tech, and it has the retention curves to prove it.
A chart published by venture capital firm Andreessen Horowitz (a16z), sourced from YipitData as of February 12, 2026, is making the rounds in tech and investor circles — and for good reason. It shows the user retention curves for four leading AI products — ChatGPT, Gemini, Perplexity, and DeepSeek — tracked over 24 months since a user’s first interaction. What it reveals has left even seasoned Silicon Valley veterans at a loss for precedent.
Garry Tan, CEO of Y Combinator, put it plainly when he saw the chart: “Does a hype cycle have a retention curve that looks like this? Literally never seen a retention cohort graph like this.” Coming from someone who has evaluated thousands of startups and their growth metrics, that’s not a throwaway comment. It’s a signal that something genuinely different is happening.
What Is a Retention Curve?
To appreciate why these curves matter, it helps to understand what they measure. A retention curve tracks the percentage of users who continue to use a product over time after their first interaction. Every cohort of new users — say, everyone who signed up in January — is followed month by month to see how many are still active. The resulting line, when plotted on a graph, tells you whether a product is keeping its users or losing them.
In almost every consumer tech product in history, retention curves follow a familiar and somewhat depressing arc: they drop sharply in the first few months as novelty fades, then flatten out at some baseline level — the “retained core” who genuinely find the product valuable. The shape looks like a ski slope that eventually levels off. The higher that flat line stabilizes, the better the product. A curve that never flattens and keeps falling is the signature of a fad.
The very best consumer apps — think Spotify, Instagram, or TikTok at their peak — might retain 30 to 40 percent of users after a year. Most apps are lucky to retain 10 to 20 percent. Products that turn out to be pure hype, driven by curiosity rather than genuine utility, often crater to near zero within six months.
What Makes These AI Curves Extraordinary
The a16z chart throws that conventional wisdom out the window — particularly for ChatGPT.
ChatGPT begins with roughly 37 percent retention in month one, a strong but not unprecedented start for a highly hyped product. What happens next is what defies the standard model. Rather than declining steeply, the curve barely dips. And then, rather than flattening, it begins to climb — steadily, persistently, month after month — reaching over 70 percent retention by month 24. The arrow on the chart points upward, because the trend is still ascending. That is an upward-sloping retention curve over two years. In the history of consumer software, that shape is essentially unheard of.
Gemini tells a similarly surprising story. It starts around 35 percent, dips slightly in the early months as many new users do, but then begins a sustained recovery and climb, reaching approximately 44 percent at the 24-month mark. That recovery — the “smile curve” shape — is itself rare. It suggests that users who stuck around long enough came to find deeper value, and that the product improved meaningfully over time in ways that pulled engagement back up.
Even Perplexity and DeepSeek, which show more conventional declining curves, retain around 14 percent of users at two years. That’s below the leaders but still respectable by the standards of consumer software.
Why This Has Never Been Seen Before
There are a few converging reasons why AI products are generating retention behavior that no prior category of software has matched.
The first is genuine, compounding utility. Most software tools do one thing. AI assistants do thousands. The longer a user engages, the more use cases they discover — drafting, coding, research, tutoring, creative work, emotional support. This expanding surface area of utility means that retention doesn’t decay as users exhaust the product’s value; it grows as they discover more of it.
The second is habit formation at an unusually deep level. AI products are increasingly embedded in daily workflows for both professionals and consumers. Unlike a photo filter app or a game, an AI assistant becomes infrastructure — something people reach for the way they reach for a search engine or a spreadsheet. Once embedded, switching costs rise and churn falls.
The third factor is that the products are getting better at a rate users can actually feel. ChatGPT in early 2023 was impressive. ChatGPT in early 2026 is dramatically more capable. Users who stayed are being rewarded with a product that keeps improving under them — an experience more analogous to a trusted colleague getting more skilled over time than to a static piece of software.
What This Means for the Industry
For investors and founders, a chart like this rewrites the valuation calculus for AI. If the leading AI products retain and deepen engagement the way this data suggests, then the revenue projections analysts have been cautious about start to look conservative rather than optimistic.
For competitors, it is a warning. The window for displacing ChatGPT on pure utility grounds is narrowing. A user base that has been deepening its engagement for 24 months is not easily lured away. Switching costs — both practical and psychological — compound over time just as the retention curves do.
For the broader tech industry, it is a moment of reckoning with the nature of this disruption. Hype cycles produce spiky adoption and rapid abandonment. They don’t produce upward-sloping two-year retention curves. What this data describes is not a bubble finding its floor — it is a category of software becoming load-bearing infrastructure in people’s lives faster than any prior technology managed to do.
Garry Tan was right to be surprised. The rest of the industry should be paying equally close attention.