AI Runs on Fuel, Not Magic: The Matatu Route Hiding Inside Every Answer

The power token ran out at exactly the wrong second. A boy had just handed over his coins at the kiosk, the owner was mid-tap on her phone, and then the little prepaid meter on the wall beeped its sad beep and the fridge went quiet. The sodas started their slow journey back to room temperature. The boy, unbothered, asked her to help him type a question into a chatbot on the cracked shop phone while they waited. She squinted at him. "You want me to ask the internet a question," she said, "but there is no electricity to keep my milk cold?"

I have been thinking about that kiosk ever since, because she accidentally said the truest thing anybody has said about AI all year. The milk and the chatbot are drinking from the same cup. We just don't see it, because our half of the cup looks like a free text box.

What actually happens when you ask AI a question

Let me walk through it the boring way, because the boring way is where the truth lives.

You type a question. It leaves your phone and travels to a building you will never visit. Inside that building are rows of specialized chips, and when your question arrives, a batch of them light up and do an enormous amount of very fast arithmetic to guess the next word, and the next, and the next. That arithmetic is not free. It pulls electricity. The chips get hot, so more electricity runs the cooling to stop them from cooking themselves. Then the answer travels back to your cracked screen and lands as a calm little paragraph, as if it strolled in from nowhere.

So the answer feels weightless to you. On the other end it was a spike on somebody's power bill, a bit of warm air pushed out of a cooling vent, and a slice of a machine that cost more than the kiosk, the shop next door, and the whole row of shops combined.

Which raises the question I cannot stop asking about anything in this business: but who controls it? Not the chatbot. The pump.

The matatu route hiding in the machine

Here is the way it clicked for me. A matatu route is not really about the driver. We like to talk about the driver, the manamba hanging off the door, the paint job, the sound system. But the route's real economics are set by three boring things: the price of fuel, the number of vehicles the stage can actually carry, and whoever controls the slots on that route.

A driver can be a genius. He can know every shortcut in the county. It does not matter if fuel doubles, or if the SACCO that owns the route only lets a certain number of vehicles run per hour. The talent is capped by the fuel and the slots, every single time.

AI is the same shape. The model is the driver, and everyone wants to talk about the driver. Faster, smarter, new version, better vibes. But the model runs on compute, and compute runs on power. Fuel is electricity. The matatu is a chip. The stage is the data center, with only so much room and only so many slots per hour. And the SACCO, the quiet people who decide who gets a slot and what the fare will be, are whoever owns the power, the chips, and the land they sit on.

Once you see that, the hype gets quieter and more interesting. The bottleneck for AI right now is not cleverness. We have plenty of clever. The bottleneck is fuel and stage space. Energy and compute.

Follow the fuel, not the demo

Watch where the real money is moving and you will notice it isn't chasing smarter chatbots. It is chasing power. Companies are signing deals for electricity years in advance. They are eyeing old power plants. They are building next to rivers and cheap grids the way a smart matatu owner parks near the fuel station with the best price. Nobody films a launch event about a substation. But the substation decides who runs the route.

So ask the SACCO question about your favorite AI tool. Who owns the stage it parks at? Who sets its fare? If the fuel price jumps, does your tool get slower, more expensive, or quietly worse, and do you even find out? A driver does not control the pump. Most AI companies do not own their power either. Somebody upstream does, and that somebody is having a very good decade.

The people who lose leverage in this story are the ones who assumed the fare would stay low forever because the text box looked free. The people who gain leverage are the ones who own fuel, chips, and the ground in between.

A small lab note

This is one reason a private notebook that runs a smaller model on your own machine is more interesting than it sounds. It is the difference between owning a bicycle for the short trips and paying matatu fare for every single errand, including the one to the shop next door. In the lab we keep circling ideas like Hapo Ndani, where the small everyday questions never leave your device and never pay route fare, and you save the long journey to the big data center for the trips that actually need a matatu. Not paranoia. Just refusing to pay the stage every time you want to cross the road.

What to do now

Next time an AI tool amazes you, do one small thing. Ask where its fuel comes from and who owns the stage. Then notice which of your daily questions genuinely need the full journey to a giant data center and which ones a smaller, closer model could answer for the cost of your own charged phone. Batch the big ones. Keep the small ones local. Care about efficiency the way a matatu owner cares about fuel, because that is the number that actually decides who eats.

The chatbot is the driver everyone applauds. But the milk and the machine drink from the same cup, and somewhere behind the calm little paragraph there is a meter beeping, and somebody you never see is holding the token.