消費者向け AI の ARPU（1 ユーザーあたり平均収益）の課題

Around a year ago, a graph of ChatGPT’s gross retention curve went viral. It was a remarkable chart: not only was the retention exceptionally high, but it also inflected upwards in a “smile”. ChatGPT users were sticking around, and many of those who weaned themselves off it found themselves coming back later.

Deedy

@deedydas

ChatGPT's product retention curves is a product manager's wet dream.

Their 1 month retention has skyrocketed from <60% 2yrs ago to an unprecedented ~90%! Youtube was best-in-class with ~85%.

6mo retention is trending to ~80%. Rapidly rising smile curve.

Generational product.

It looked pretty clear by extrapolating this retention curve, consumer AI was going to be the biggest market on earth.

But we all missed something at the time: this was measuring gross retention, not net. Even the most engaged consumer users were still only paying $20 per month. All the value from increased usage was being kept by the user, not OpenAI.

$44 billion of Anthropic B2B revenue and a slow plateauing of ChatGPT consumer bookings later, sentiment has shifted. B2B, and specifically coding, is where the action is from a research, fundraising, and sentiment perspective.

Consumer AI has stalled, and enterprise has exploded.

That’s what this post is about: why has consumer AI monetised so poorly in comparison to its societal impact, and what this means for OpenAI, Anthropic, and everyone else.

Products make money when they either do something useful for their customers (“create value”), or displace something they’re already paying for (“capture value”).

For coding agents, they evidently do both: Claude helps you build something previously beyond your capabilities (create value), and potentially mean you don’t need to hire an additional engineer (capture value). Legal AI does the same: Claude can summarise 100-page contracts in seconds, and in doing so, replace thousands of dollars worth of overtired associate-hours.

So in a B2B context, a huge amount of valuable (or in many cases pointless, but still expensive) work is already being done, and AI can be used either to do more work, or use fewer humans to do the same work.

In other words, the incremental value of B2B AI is very easy to monetise.

This has shown in the numbers.

Anthropic’s 3x revenue growth in the last quarter is not because the number of developers has increased proportionately; it’s because each user is spending dramatically more every month.

Unfortunately, this is not the case for consumer AI.

For most consumers, “creating value” isn’t something they are looking for in their technology choices. And they are already incurring spend that can be “captured”.

If ChatGPT gives them a quick answer to a question that was bugging them, most people don’t see that as fundamentally valuable - they could have Googled it anyway! It saved them time, but it didn’t in any way make them money, so why should they pay for it? And if they make a fun image on ChatGPT Image-2 that they use as their PFP for a few days, can they look themselves in the mirror and tell themselves it’s really worth paying for?

Then there’s value capture. If ChatGPT helps them fix their boiler so they no longer need to pay for the plumber to come, they’ve saved some money, but why does that mean they need to pay ChatGPT? Consumers have become so accustomed to being given free value on the internet that many feel indignant at even being asked to cough up $20 per month.

In other words, consumers don’t expect to have to pay for the incremental value they capture above their monthly subscription; they expect to keep it.

In a sense, ChatGPT’s consumer business has been a victim of its own rapid success.

ChatGPT’s user growth is no longer exponential and has stalled before reaching 1 billion, mostly because there are not enough affluent users around the world (outside China) to keep growing fast.

This is the reason OpenAI has moved to their

of ads. I have previously expressed skepticism over ChatGPT’s ad businesses to scale at the rates required for them to make a dent in ChatGPT’s current valuation. The issue is the same as their consumer subscription business: ad business CPMs (equivalent to ARPU) are very low outside the very highest-converting ad units (search), which aren’t possible within their current vision for ChatGPT. Social ads are a big business, but the individual ads are very cheap; the scale of the business comes from ad load (and time spent) in Meta and TikTok’s platforms, which is nowhere near replicated in LLMs.

Sasha Kaletsky

@SashaKaletsky

People are underestimating how hard it will be for ChatGPT to integrate ads.

They have:

• No obvious native ad unit

• Limited commercial intent usage

• Trust built on objectivity

• Competitors without ads

They're the best (ex-Meta) product people in the world so they have a

With ChatGPT, OpenAI has driven the most world-changing technology in my adult lifetime. But it looks like, similarly to Amazon, Microsoft and NVIDIA before it, its consumer business's financial impact could be dwarfed by the enterprise business that results from it.

ChatGPT is not alone in this ARPU trap. Many other consumer AI businesses facing TAM exhaustion are facing something similar, and look at the exponential upsell growth from their B2B peers with envy.

For standalone consumer AI platforms to grow to

+ in value, they will need to figure out how to charge their users much, much more.

And none of the current monetisation options are going to do that.