The ARPU Problem of Consumer AI

At one point, ChatGPT's gross retention curve spread out in a smile shape, making it look like consumer AI would become a massive market. But that metric was only gross retention, not a net figure. Even as users used it more, spending actually stayed around $20 per month, and the increased value stayed with the users rather than going to OpenAI.

A product makes money when it creates new value for customers or replaces spending they were already making. B2B AI can satisfy both of these conditions relatively easily. Coding agents make possible things that were previously impossible, and legal AI replaces lawyer time worth thousands of dollars. In other words, value creation and value capture happen simultaneously, and Anthropic's 3x quarterly revenue growth also came from increased monthly spend per user rather than from user count.

• Even though consumer AI's answers and image generation are useful, many users don't see them as something worth paying for.
• Even if it saves them the cost of calling a plumber, users don't think AI should get to keep those savings, and having grown accustomed to the internet's free value, they resist subscription fees.

This structure creates a ceiling on ARPU for consumer AI.

Globally—especially outside China—there aren't enough high-income users, so ChatGPT's consumer growth slowed before reaching a billion users, and OpenAI has come to consider advertising as a last resort. But ads inside an LLM are hard to make convert as highly as search does, and with low CPM, it's difficult to replicate the ad density and dwell-time models of Meta or TikTok.

In the end, consumer AI ends up making relatively little money compared to its social impact, while actual revenue and market value are likely to be captured by the enterprise side. Just as Amazon, Microsoft, and Nvidia did, OpenAI too could make more money from the enterprise businesses that its technology gives rise to than from the consumer technology itself—but for an independent consumer AI platform to reach a $1T+ valuation, it needs to get users to pay much more. With the current subscription and advertising models, that answer isn't in sight.