| url | https://andrewchen.com/end-of-the-billion-user-startup/ |
|---|---|
| raw | raw/highlights-end-billion-user.json |
TL;DR: The era of billion-user ad-supported consumer startups is over. Growth channels are saturated, viral mechanics have been nerfed, and the economics no longer pencil out. The new playbook looks like Duolingo: single-user-utility + game mechanics. The product has to be valuable on day one with zero other users. Then ChatGPT happened and immediately broke the rule for the obvious-in-retrospect reason that the original argument didn’t anticipate.
What it means
Chen’s argument is that the distribution channels that created Facebook, Instagram, and WhatsApp have been closed. Viral growth through contact importers, open social graphs, and cheap mobile acquisition has been regulated, rate-limited, or arbitraged to zero. Building the next billion-user social network through the old playbook is no longer feasible. The math doesn’t work, the channels don’t exist, and the platforms (Apple, Google, Meta) actively close any new viral loop within months of it appearing.
The replacement model looks more like Duolingo: an app with strong single-user-utility (you can learn a language alone, today, with no friends on the platform) combined with game mechanics (streaks, leaderboards, XP) that create stickiness without requiring a massive social graph. The product is valuable on day one with zero other users, and the engagement loops are built into the core experience rather than depending on network-effects that require a critical mass nobody can afford to acquire anymore.
This has big implications for how startups think about moats. If you can’t realistically build a billion-user network, the network-effects playbook becomes less central. Instead, moats come from habit formation, switching-costs through accumulated user data and progress, and branding built through consistent quality. The post-2020 winners look very different from the 2010s winners, and the playbook diverged hard.
The argument
Growth channels are saturated. Every major consumer acquisition channel — App Store optimization, Facebook ads, viral invites, SEO — has been arbitraged to near-zero marginal returns. The cost of acquiring a consumer user has risen to the point where ad-supported models can’t make the math work for new entrants. The companies that look like they’re “growing through TikTok” are mostly drafting on a single platform-specific quirk that will close within 12 months.
Network effects can be scoped. You don’t need all your friends on a platform to get value from it. Dating apps, gaming communities, and interest-based networks build network-effects around activities rather than social graphs. This means you can reach critical mass in a niche without needing universal adoption — and the niche-network strategy is much more durable than the global-network strategy used to be.
Game mechanics as engagement moat. Duolingo’s streaks, XP systems, and leaderboards create retention without social dependencies. These mechanics produce a new kind of switching-costs — losing a 500-day streak feels like losing something real, even though the “asset” is entirely artificial. This is a new kind of lock-in that doesn’t require network-effects to function.
Single-user utility is the foundation. The bootstrapping lesson from status-as-a-service applies here too: the product must work for one person first. But Chen goes further — in many cases, single-user utility isn’t just the bootstrap; it’s the entire value proposition. The social layer is optional, not essential. Once you accept that, the design space opens up enormously.
The ChatGPT counterexample
Chen wrote this before ChatGPT proved every part of the argument wrong on its own peculiar terms — by reaching ~100M users in 60 days through pure organic word-of-mouth and a one-click signup. The interesting question is whether ChatGPT is the exception or the new rule. The honest answer is probably “exception”: ChatGPT’s growth was driven by a once-in-a-decade demand release (the world realizing AI worked), not a sustainable channel. The next 300M users are going to be much harder than the first 300M, and the playbook for getting them is going to look more like Chen’s “single-user utility plus game mechanics” framing than like the Facebook viral-loop playbook of the 2010s. Chen’s framework still describes the steady state. ChatGPT is the discontinuity that lives outside the framework’s domain.