The Intelligence Bill

I want to put a number on the record so it can be checked. By 2030, one billion people will pay $100 or more per month to OpenAI or Anthropic. That is $1.2 trillion a year in gross revenue between two companies. By 2040, three billion people will pay $200 a month, which is $7.2 trillion a year. I think both numbers are low, and the rest of this piece is about why.

• 20301,000,000,000 × $100 × 12$1.2 trillion / yrRoughly what every mobile operator on earth billed in 2025, combined.
• 20403,000,000,000 × $200 × 12$7.2 trillion / yrAbout the same as the entire mobile ecosystem’s contribution to world GDP in 2025, per GSMA.

For scale, the two companies together are on a run-rate of somewhere around $90 to $100 billion right now. Anthropic disclosed a $47 billion run-rate when it closed its Series H on 28 May 2026, and a figure of $65 billion was reported for late July. OpenAI has not disclosed a comparable current number; reporting from early 2026 put it above $25 billion annualized, and the company’s own plan, reported by Bloomberg in February, has revenue topping $280 billion in 2030. So I am predicting roughly four times OpenAI’s own 2030 forecast for the pair, and I am calling that conservative. I owe an argument.

Here is how OpenAI models its own consumer future, according to documents The Information reported on in February 2026. It projected 2.6 billion weekly users by 2030, of whom 8.5 percent would pay for a Plus plan. That is about 220 million subscribers. The comparators the reporting reached for were Spotify, at 281 million subscribers, and Netflix, at 302 million. At the time, OpenAI had about 800 million weekly users and about 35 million paying for Plus or Pro.

Look at the shape of that model. It is a media business. A free product used by billions, a single-digit conversion rate, a $20 price point, and advertising to monetize the rest. OpenAI’s plan reportedly counts on $100 billion of ad revenue by 2030. That is the economics of a streaming service or a social network. It assumes that what people are buying is content and convenience, and that most of them will happily take the free version.

A chatbot is media. Something that works for you while you sleep is a utility. Nobody prices a utility like a streaming service.

I think the model is wrong about the product, not the arithmetic. The Plus plan they are modeling is a chat window. You open it, ask it something, close it. Conversion rates for that will always look like Spotify’s, because that is what Spotify sells: something you open when you want it. The product that actually matters is the one where the intelligence runs without you opening anything.

That is the thesis in one scene. The baseline condition for a functioning adult is about to include doing very smart things all the time, continuously, without attending to them. Not asking a chatbot clever questions. Having intelligence that runs on a schedule and on triggers, on your behalf, whether or not you remember it is there. I teach this as ambient intelligence: invisible, always on, never prompted. In practical terms today it means scheduled tasks. Both companies already ship the primitive: Anthropic’s scheduled tasks and routines in Claude, OpenAI’s Tasks in ChatGPT. Today they are rough. By 2030 they will be the product.

Once that exists, the choice is not whether to pay. It is whether to fall behind the people who do. The person with fifty quiet things running for them at all times is not slightly better off than the person with none. They live in a different regime. They respond faster, forget less, catch more, and make fewer unforced errors, every day, compounding. The gap is not a feature gap. It is the gap between someone with a phone and someone without one in 2015.

I keep coming back to that comparison because it is the honest one. In 2014 there were about a billion smartphone users. By 2026, estimates put the number around 5.6 billion, and GSMA counts more than 5.9 billion people with a mobile subscription. The world found the money for a phone bill, at every income level, because being unreachable stopped being a viable way to live. Being un-augmented is about to stop being a viable way to live in exactly the same way.

The $200 tier already exists. OpenAI created it in December 2024; Anthropic followed with the $100 and $200 Max plans in April 2025; by August 2026 five providers sell a $200 tier. Those tiers were priced for heavy chat users. That is what makes them look extravagant. Priced instead as the cost of intelligence running in the background all day, doing what would otherwise take hours of your own time, $100 is not a premium. It is the cheapest hour anyone has ever bought back.

Think about what a monthly bill of $100 is competing against. Not a Netflix subscription. It competes against the hours of your own life you currently spend remembering, checking, chasing, and filing. The people paying it will not experience it as a software purchase. They will experience it the way they experience the phone bill: nobody thinks of their phone as a business expense, even though it is the device they run their whole life on. The bill is the price of being present.

Can a billion people afford it? Brookings projects that 5.6 billion people will be middle class or wealthier by 2030 on its $11-a-day-and-up definition. The top billion of the world by income can pay $100 a month without discussion, and a large share of them will have it paid by an employer, the way phones and laptops are. The harder question is the next two billion, which is why my 2040 number is the one I would defend less fiercely. But note that $100 a month is what a real assistant costs per day in most rich countries. The price is not the obstacle. The obstacle is the product being good enough that skipping it visibly costs you. That is a 2027 problem, not a 2030 problem.

A forecast is worth nothing without the conditions under which it fails. Here are both claims with what is already true, what still has to happen, and where I am most likely wrong.

Claim one: one billion people at $100 a month by 2030

Already true

Roughly a billion people use ChatGPT every week. Two labs are at a combined run-rate near $100 billion less than four years after the first consumer product launched. The $100 and $200 tiers exist and sell. Scheduled, autonomous execution exists in both products.

What has to happen

Scheduled intelligence has to go from a feature to the default way people use these products, so that the free chat window looks like the demo and the paid always-on layer looks like the product. Reliability has to reach the point where handing it real authority is unremarkable. And compute has to keep up; grid connections, transformers, and turbines are already the binding constraint, with multi-year delays reported on new data-center projects.

Where I am probably wrong

The bill might not go to OpenAI or Anthropic. If Apple, Google, Microsoft, or a carrier bundles the always-on layer into a device or a plan, a billion people will pay $100 a month for intelligence and it will show up as somebody else’s revenue, with the labs paid wholesale. I would still be right about the bill and wrong about the payee. The second risk is price: competition and open-weight models could push the working tier toward $20 while the $100 tier stays a specialist product. Then the users are right and the revenue is off by five.

Claim two: three billion people at $200 a month by 2040

Already true

The phone bill proves three billion people at every income level will pay a recurring fee for a capability once life without it becomes untenable. The global middle class is on track to exceed five billion before 2030.

What has to happen

The always-on layer has to become as necessary as connectivity: employers assume it, governments interact through it, and the person without one is visibly disadvantaged. Prices have to rise with capability rather than fall with commoditization, which means the product has to keep absorbing higher-value work rather than plateauing.

Where I am probably wrong

Fourteen years is long enough for the whole category to restructure. Intelligence could become a public utility, priced by governments, or so cheap that $200 buys something unrecognizable. My $200 figure assumes the value keeps rising and the price tracks it. If cost per unit of intelligence keeps falling faster than the amount of intelligence a person needs, the bill shrinks even as the dependence grows. That is the failure I would bet on if I had to bet against myself.

That is the whole argument. The industry is pricing a chatbot, and a chatbot deserves a chatbot’s conversion rate. The thing people will actually be paying for is ambient intelligence: invisible, always on, never prompted. That gets paid like a utility bill, and the phone bill is already $1.2 trillion a year.