<100 subscribers
Share Dialog
Let’s play a game. You, a bright-eyed entrepreneur, have a world-changing idea for an AI application. Maybe it’s a chatbot that finally tells you which of your plants are judging you, or an AI that can generate the perfect excuse to get out of a boring meeting. You fire up your laptop, ready to code your way to a billion-dollar valuation.
Congratulations! You’ve just entered the Application Layer. Welcome to the kiddie pool of the AI ocean. It’s warm, crowded, and everyone is splashing around with the same inflatable toys (looking at you, OpenAI API). The barrier to entry is a monthly subscription fee and a strong coffee habit. Competition is fiercer than a seagull fighting for a single french fry. But it’s fun! You’re building!
But have you ever stopped to wonder what’s powering your little slice of genius? Let’s take a dive into the deep end, where the real monsters—and the real money—live. It’s a layered world of silicon, sweat, and government regulations, and it looks less like a bake-off and more like a fight over who owns the only oven on Earth.
The Application Layer: The Cupcake Bakers
This is us. We’re the bakers using a pre-made mix (an LLM) from the store (OpenAI, Anthropic, Deepseek, etc.) to whip up delightful cupcakes (apps). We compete on frosting design and sprinkles. It’s chaotic and wonderful. But we’re all nervously glancing at the store, waiting for them to suddenly start selling cupcakes that are cheaper, prettier, and wittier than ours. The store, it turns out, is run by wizards.
The LLM Layer: The Wizard Chefs
These are the Metas and Googles of the world. They’re not baking cupcakes; they’re inventing flour from scratch. To train one of their models, they need to perform a number of calculations so astronomically large it has no name—let’s call it a “gazillion-flop.” Doing this on your gaming PC would take roughly the time between the invention of the wheel and the invention of the selfie stick.
So how do they do it? They throw money at the problem until it surrenders. Imagine needing to bake a cake for every person on Earth, simultaneously. Their solution? Buy every oven on the continent. Meta used 16,000 of Nvidia’s H100 GPUs—the Ferrari of ovens, costing $40,000 each—just to train Llama. That’s over $640 million in ovens, and that’s before the electricity bill. The barrier to entry here isn’t high; it’s in orbit.
The Infrastructure Layer: The Oven Manufacturers
China’s response? “Fine. We’ll bake with Easy-Bake Ovens.” Companies like DeepSeek are performing culinary miracles with the H800, the international, nerfed version of the H100. They used 2,048 of them to create DeepSeek V3—a model that went toe-to-toe with OpenAI’s best. It’s the equivalent of winning a Michelin star using only a toaster and a hairdryer. If you are the US its quite maddening, for the rest of us Its quite impressive.
The Production & Supply Layer: Making the Ovens… and the Flour
This is where it gets really spicy. You can’t make a top-tier oven (GPU) without incredibly advanced semiconductors (the flour). Making this “flour” requires something called a “fab”—a factory so absurdly precise and clean it makes an operating room look like a mud-wrestling pit.
These fabs are logistical nightmares that cost more than the GDP of a small nation. The most advanced flour is currently milled almost exclusively in Taiwan by a company called TSMC. China’s “Made in China 2025” plan is essentially their declaration: “We will build our own flour mill!”
But there’s a catch. To make the finest flour, you need a magical mill called an EUV machine. There is essentially only one company in the world that makes these fairy-tale devices: ASML in the Netherlands. And guess what? The US has (politely? forcefully?) asked them not to sell the magical mill to China.
So China’s barrier isn’t just money or talent (they have more STEM graduates than anyone); it’s being locked out of the only store that sells the magic wand.
The Electricity Layer: Who Owns the Sun?
And beneath it all, humming ominously, is the final layer: power. All these ovens and magical mills are insatiably hungry. The energy needed to run this AI race is so vast that companies are literally searching for entirely new sources of electricity. We’re not just building AI; we’re trying to invent the sun to power it. The next big breakthrough in AI might not be an algorithm, but a fusion reactor.
The African (and Global) Dilemma: Do We Bake Cupcakes or Build an Oven?
For now, Africa’s most immediate opportunity is two-fold: be the best darn cupcake bakers (application layer) in the world, solving local problems with these powerful, rented tools, and provide the raw materials (metals, minerals) needed for all those ovens and magical mills.
But the window for cupcake dominance might be closing. The wizards in the LLM layer are racing towards AGI—Artificial General Intelligence. This is the dreaded day when the store not only sells the best cupcake mix but also develops sentient, superior taste buds and starts baking itself, cutting us out of the loop entirely.
So, to the lawmakers of the world, as you hastily draft AI policies between committee meetings and lunch: understand the layers. Are you stifling your cupcake bakers with red tape while other countries corner the oven market? Are you investing in the electrical grid that will be needed to power the next decade?
The AI race isn’t a single sprint. It’s a relay race through an obstacle course where each lap depends on the one before it. And right now, everyone is desperately trying to run the next lap while also trying to steal the other team’s shoes. It’s chaotic, it’s hilarious, and it will define the future of power on this planet. Now, if you’ll excuse me, I have a cupcake to frost before the ovens become self-aware.
Fabian Owuor
