Sam Altman has said it plainly, more than once. Intelligence, he told a room of bankers at the U.S. Federal Reserve in July 2025, is on track to become “too cheap to meter,” a phrase borrowed from 1950s nuclear power hype that never actually came true for electricity either. In June 2024 he’d already framed it the same way on record: intelligence as a utility, priced per token, delivered like power to a grid. Read that sentence again. The most powerful man in AI is not describing a tool. He is describing a metered dependency, and he is the one holding the meter.
The Business Model Is the Message
OpenAI sells tokens, the units its models use to think and respond. Every “gpt-4o-mini advancing cost-efficient intelligence” announcement, every price drop per million tokens, is not charity. It is a company perfecting the art of making thinking itself a billable commodity, one you rent rather than own. Altman’s own blog admits the plan: as data-center production automates, “the cost of intelligence should eventually converge to near the cost of electricity.” That means intelligence is being deliberately engineered into infrastructure you cannot unplug from, the same way you cannot opt out of the power grid once your whole life runs on it.
What the Science Actually Shows
This wouldn’t matter as much if outsourcing your thinking were free of cost. It isn’t. In June 2025, MIT’s Media Lab published “Your Brain on ChatGPT,” an EEG study tracking 54 participants writing essays using either ChatGPT, a search engine, or nothing at all. The finding: ChatGPT users showed the weakest, least distributed brain connectivity of the three groups, alongside more homogeneous, less original writing that human teachers could reliably identify as AI-shaped. Lead researcher Nataliya Kosmyna and colleagues coined a term for it: “cognitive debt,” a decline in independent reasoning that persisted even in a later session when the same participants were asked to write without AI help. A separate 2025 CHI conference study of knowledge workers found self-reported reductions in critical thinking effort tied directly to how much they trusted the AI output. This is not fringe alarmism. It is peer-reviewed and MIT-branded, even as researchers caution the sample was small and the study remains a preprint.
Rules for Thee, Not for Me
Now hold that against how AI’s own architects raise their children. Peter Thiel, PayPal cofounder and early Facebook investor, told the 2024 Aspen Ideas Festival he limits his kids to ninety minutes of screen time a week, a figure that reportedly drew audible gasps from the audience. Bill Gates didn’t give his children phones until age fourteen. Steve Jobs restricted iPad use at his own dinner table. The Waldorf School of the Peninsula, two miles from Google’s headquarters, bans screens for children under eleven and markets a tuition of nearly $47,500 a year around “real childhood,” while several of its own board members work professionally in AI. Thiel put the quiet part out loud himself: “There are probably a lot of people in tech who do something quite similar for their own families. There’s some questions that that might lead you to ask.” To be fair, this pattern isn’t universal, some tech parents favor moderation over prohibition, and outlets have pushed back on treating one private school as proof of a grand conspiracy. But the pattern among the industry’s most senior architects is documented, repeated, and telling on its own terms.
What This Means for Africa
Now bring it home. Africa hosts less than 1% of global data-centre capacity, and an estimated 80% of the continent’s internet traffic still routes through foreign servers, mostly in Europe. Kenya’s flagship $1 billion Microsoft-G42 data centre deal, brokered by Washington in 2024 to counter China’s influence, had still not broken ground by August 2025; President William Ruto himself admitted powering it would require shutting off electricity for half the country. Nigeria’s grid delivers roughly four hours of reliable power a day, forcing data centres onto diesel generators. Meanwhile Nigeria, Egypt, and Kenya have all published draft AI strategies since January 2025 naming dependence on Google, Microsoft, Nvidia, and Meta as a direct threat to national security.
This is the real danger: African governments, students, and creatives are being sold “the future” by companies that don’t let their own children touch the product casually, built on servers Africa doesn’t own, running on power Africa doesn’t reliably have, governed by regulation Africa hasn’t finished writing. Selling intelligence as a utility only works if the customer never builds their own grid. The question isn’t whether AI is powerful. It’s who gets to own the meter, and who’s left paying whatever it reads.
