Every argument about AI starts with a number almost nobody checks: how many people actually use it. I checked it twice, against independent sources, before writing this. The number is 17.8% of the world's working-age population as of Q1 2026, and it moved 1.5 points in that quarter alone. Run that pace flat for 5 years, zero acceleration, and you land near half the working-age planet. The forces underneath it aren't flat.
Adoption has always had 3 gates: a connection, a device, and a price you can clear. For 20 years those gates opened one at a time, rich world first. Right now all 3 are collapsing at once, fastest in the places that were last in line.
The connection gate. 2.17 billion people are still offline, mostly across Southern Asia and Sub-Saharan Africa, because towers and fiber never penciled out where they live. Direct-to-cell satellites delete that math. Coverage stops needing hardware on the ground. There are 650+ direct-to-cell satellites overhead right now, 400 million people already sit in the footprint across 22 countries, and Airtel's deal queues up 174 million African subscribers starting this year. The marginal cost of covering one more village just went to zero.
The device gate. Already open, and almost nobody frames it this way. Smartphones are 89% of the handsets in use on Earth, and 5.83 billion people carry a phone. The sub-$50 Android closed this gate over the last decade while everyone was watching something else.
The price gate. The steepest curve in the story. Matching what GPT-4 could do in March 2023 cost $30 per million tokens then. It costs under $0.50 on open-weight models now. That's 60x cheaper in 3 years, and at that price free is the business model: Meta ships AI inside a messaging app with 3 billion users, and OpenAI's roughly $4.50 tier helped push India to 100 million weekly ChatGPT users. The proof is already on the board: India, on free and near-free tiers, is Meta AI's largest market and ChatGPT's second largest.
History already showed us what happens when gates open where there's no incumbent system to displace. Africa skipped landlines and went straight to mobile. It skipped bank branches and went straight to M-Pesa. AI is set up to skip the desktop era entirely: voice-first, local language, inside apps people already use. The last 2 big adoption jumps came from models getting good at Korean and Thai. Hindi, Swahili, Bahasa and Yoruba haven't had their jump yet. When they do, whole populations flip at once, because language was the last closed gate.
So read the projection fan straight. The bear case, where satellites slip and data stays unaffordable, still doubles today's number by 2031. The base case just extends the measured pace. The fast case only needs things already on the calendar to happen on time. The next 4 billion aren't hypothetical. They're in line. The question that actually matters, and it's the whole subject of part 2, is who owns the pipes they arrive through.