One person launches swarms of AI agents, programs built to act on their own. Another has never touched an AI model. The difference between the two is measured in several exponents, a scale where each step multiplies the one before. Sorting the world by access to AI shows the tiers between them, and each tier up holds a far smaller group of people, from billions down to thousands.
The sorting starts with the whole world, the base for every other count, where about 8.3 billion people live in 2026. A tier here is a level of access to AI, and each level brings a different set of models. From the bottom, the tiers are no AI, free, paid and the very top. An estimated one to two billion people have used some version of AI, usually a free one running on the least capable models. Each count is a rough estimate, but the order of the tiers holds across the whole range of each one.
Above the free tier, the paid tier holds people inside companies that are adopting AI, who see its benefits across their daily work. An estimated 10 million to 50 million people belong to this group, and the true count might be lower. Even the top of that range is a small fraction of the world.
The top tier is the smallest. About 10,000 people work at the organizations building the most advanced AI, known as frontier labs. They have unrestricted access to AI models and large amounts of computing power, the processing capacity AI runs on. No one else has this level of access, and every tier below faces limits on the models, the computing power, or both.
Computing power changes what one person can get done. It is what lets a frontier lab employee run those swarms, large groups of agents working at the same time. The agents take on long-running tasks, jobs they keep working on without a person stepping in at each point.
The concern is how many people will be left behind because they cannot reach AI models the way others do. Several billion people have no AI access at all.
