On the consumer side, I estimate there are somewhere between 75m and 100m monthly paid subs across ChatGPT, Gemini, Claude, and Grok. That compares to a global daily internet audience of about 5.5B. Over the past nine months, we’ve seen model providers increasingly experiment with a wider range of tiered pricing. For example, Grok, ChatGPT, and Claude each offer about five options.

• ChatGPT: $0, $8, $20, $100, and $200.
• Gemini: $0, $5, $20, $100, and $200.
• Claude: $0, $20, $100, and $200.
• Grok: $0, $10, $30, $100, and $300.

The fact that they are trying so many options underscores a belief inside these companies that both the personalized AI and business use cases are just forming. For example, all four of the majors listed above offer a paid range basically between $10 and $200 a month. That wide gap speaks to the willingness of a small group to pay a lot per month because they’re aggressively using the tools. On the flip side, the typical user who pays $10-plus a month is just thinking, “I want my chatbot to have access to the most advanced model.” Down the road, as the nascent user increases their usage, driven in large part by the growth of agents, they’ll increasingly hit token limits. That will trigger conversion from free to paid, and from paid to paying more each month.

Five years from now, I expect the low-end options to remain largely unchanged and account for 95% of paid subs and 50% of total subscription revenue. On the high end, I expect power users to account for 5% of total paid users and 50% of revenue. While that math may be hard to believe, it’s based on an expectation that the average paid personalized AI user will spend $15 a month compared to power users spending $300 a month. In other words, this is a rare case when the 80-20 rule doesn’t apply.

The race to be a winner in personalized AI is at the top of the list for the megacap tech companies, as evidenced by Zuckerberg’s recent Meta Connect keynote, where he described Muse as the centerpiece of what Meta is building. That begs the question: What is at stake when it comes to the revenue potential around personalized AI? Let’s look at the two largest consumer platforms, Apple and Meta, as starting points for sensitivity to the numbers.

Apple: I estimate Apple has about 1.67B customers. That implies that each customer has 1.5 Apple devices. So, with over 2.5B active devices and 1.5 devices per user, that yields 1.67B users. My conservative case is that 20% of those become paid users at $10 a month to get access to more tokens or additional features. That would add about $40B a year to revenue and $32B to operating income (80% operating margin), or 18% to operating income. My aggressive case is that 50% pay $20 a month. That would add about $200B to revenue and increase operating income by about 90%.

Meta: My conservative case is that 5% of Meta’s 3.6B daily users pay $5 a month. That would add $11B a year to revenue and about $9B to operating income, or increase operating income by 8%. My aggressive case is that 25% pay $10 a month, which would add $108B to annual revenue and $86B to operating income, or increase operating income by 80%. Note that Meta’s Muse overall revenue opportunity exceeds this sensitivity, given they’ll also have a transaction model where Muse takes a cut of the transaction value. I believe this new revenue bucket would more than offset the headwind that the company’s traditional ad business will face.

In the end, these sensitivities are based on a simple question: Do you believe that personalized AI will blow us away? My sense is the early products, Muse, Grok Bot, and even the new Siri, are doing just that. Think about how good these bots will be in five years and all that we can get done with them. If that is the future, companies like Apple, Meta, Google, and SpaceX will find ways to withhold just enough features to trigger a measurable number of users into paid subs.

On the enterprise side, Ramp’s AI Index puts average monthly AI spend in the U.S. per employee at $13. That number includes LLM subscriptions, coding-agent subscriptions, API tokens, and GPU cloud and infrastructure, and it has barely lifted off the floor since early 2024. Using software spend as a proxy for what AI spend will eventually be, we see a large gap with the average monthly spend of $780 (source: Vertice). While I don’t have a direct sensitivity that maps to the consumer one above, I note that the gap between AI spend and what enterprises are used to paying per month for software is a sign that there is measurable wallet share to be gained.

Disclaimer