The Text Box and the $690 Billion
April 5, 2026I built something last month that would have taken a team of five about three months. It took me a weekend. One person, one prompt, one laptop.
The next Monday, I read that Microsoft, Google, Amazon, and Meta are collectively spending $690 billion on AI infrastructure this year. That's not a typo. Six hundred and ninety billion dollars. More than the GDP of most countries. Spent by four companies to build machines that let me — one person — do what fifty people did.
Both of these things are true at the same time. And that's the part no one is talking about honestly.
The paradox
AI is the most concentrated technology ever built. And simultaneously, the most democratizing.
On one end: Nvidia controls 85% of AI chip revenue. One company. Its software ecosystem, CUDA, isn't just a product — it's the alphabet in which all of machine learning is written. You don't compete with the alphabet. You write in it. The chips are manufactured by one company (TSMC), using memory from three suppliers. The models run in data centers owned by four hyperscalers. The supply chain looks like a funnel with one exit.
On the other end: a solo founder named Maor Shlomo built an AI app builder alone, hit 300,000 users in six months, and sold it for $80 million. Sam Altman has a group chat with tech CEOs betting on when the first one-person billion-dollar company will happen. Dario Amodei puts the odds at 70-80% it happens this year.
The means of production are now a text box. And that text box sits on top of a $690 billion infrastructure stack.
Both statements are true. That's not a contradiction. That's the new physics.
This has happened before (sort of)
Standard Oil controlled 95% of American oil refining by 1880. Here's the part people forget: it lowered the price of kerosene for consumers while crushing every competitor. The monopolist didn't raise prices. He made the product so cheap that alternatives became pointless.
Sound familiar? ChatGPT is free. Claude is free. The most powerful reasoning engines in human history cost nothing to use. While the companies behind them burn billions — OpenAI alone will burn $17 billion this year and won't turn a profit until 2030.
Electricity followed the same pattern. Samuel Insull, Edison's protégé, built a power monopoly by lowering prices and expanding access. Railroads concentrated, then became regulated utilities. Every infrastructure technology follows the same sequence: concentrate, democratize access on the monopolist's terms, then maybe — decades later — regulate.
But here's where the pattern breaks. Oil took 30 years to concentrate and another 30 to break up. Electricity took half a century to become a utility. AI is doing both — concentrating and democratizing — in the same year. In the same quarter. In the same product.
The speed is the difference. And nobody has a playbook for speed like this.
The Jevons engine
In 1865, William Stanley Jevons noticed something that still breaks people's intuitions: when steam engines became more efficient, coal consumption increased. Not decreased. Efficiency didn't reduce demand. It expanded the market.
AI is running the Jevons Paradox at scale. When DeepSeek trained a competitive model for $5.6 million — a fraction of what OpenAI spent — Nvidia lost $600 billion in market cap in a single day. "Cheaper AI means less infrastructure needed," the market said. Then, within days, Meta raised its AI spending to $65 billion. Cheaper AI didn't reduce spending. It made everyone spend more. Because when something gets cheaper, more people use it. And when more people use it, you need more of the thing that was supposed to be less necessary.
This is the engine inside the paradox. Democratization and concentration aren't opposing forces. They're the same force viewed from different ends of the value chain. Cheaper access expands demand. Expanded demand requires more infrastructure. More infrastructure concentrates power.
The text box gets more democratic. The server farm gets more concentrated. And both accelerate together.
Who disappears
The question everyone asks is: who loses?
The answer isn't workers or companies. It's the middle.
The barbell is forming. On one end: solo founders, tiny teams, individuals with AI leverage doing what departments used to. On the other: hyperscalers, the four or five companies with enough capital to build the infrastructure. The middle — the 50-person startup, the mid-sized agency, the regional IT firm — has nowhere to go. Too big to move like an individual. Too small to build like a giant.
This isn't just companies. It's the same shape as the workforce. The top 1% of engineers — building models, infrastructure, novel systems — become more valuable than ever. The middle 60% — building standard apps, integrations, CRUD — become replaceable by a prompt. It's not that coding dies. It's that average coding dies.
Young tech workers between 20 and 30 have seen unemployment rise by almost three percentage points since early 2025. Entire economies built on outsourcing — India, Philippines — are watching their middle-class growth engine stall. Goldman Sachs estimates 300 million jobs affected globally.
The losers aren't the people at the bottom. They were already struggling. The losers are the people who thought they were safe.
The philosopher that isn't one
Here's the part that keeps me up.
AI doesn't do philosophy. But it does something no human philosopher can: it holds contradictions without flinching.
Marx and Adam Smith in the same context window. Buddhist economics and Hayek. The Gita and game theory. A human philosopher picks a tradition and spends a career defending it. An AI holds all of them simultaneously, without the cognitive dissonance that makes humans choose sides.
This isn't synthesis. It's something stranger. AI is a philosophical particle accelerator — smashing traditions together and seeing what new particles emerge.
Marx never imagined means of production that cost zero to replicate. Smith never imagined markets where the marginal cost of production approaches zero. Buddhist economics never imagined desire being manufactured algorithmically at scale. Every philosophical tradition was built for conditions that AI has already invalidated.
And the new conditions don't fit any single tradition. They need something that doesn't exist yet.
I don't think AI will create that philosophy. But I think it's forcing its creation. By making the old frameworks visibly insufficient. By making the contradictions impossible to ignore.
We went from an attention economy — your eyeballs were the product — to a cognition economy. Your thought patterns are the product now. Not what you look at, but how you think. Every preference click, every prompt, every interaction is training data. The commodity isn't your attention anymore. It's your mind.
No philosopher anticipated that. No economic theory accounts for it. We're in uncharted territory and we're moving at Jevons speed.
The thing I can't resolve
Ronald Coase won a Nobel Prize for explaining why companies exist: because coordinating work through markets is expensive, so firms form to reduce transaction costs.
AI is eliminating those transaction costs. The one-person unicorn isn't about one person being superhuman. It's about making the corporation unnecessary for the kind of coordination that used to require one. When coordination is free, the firm dissolves.
But the infrastructure that makes coordination free? That requires the largest firms in human history. The companies spending $690 billion. The ones building data centers that consume a quarter of Virginia's energy.
So we need the biggest companies ever to exist in order to make companies unnecessary.
I don't have a resolution for that. I don't think one exists yet. The paradox isn't a bug. It's the operating system.
Capitalism produced the most powerful tool in its history. And that tool is making capitalism incoherent. The system requires scarcity; AI produces abundance. The system requires wage labor; AI eliminates it. The system requires competitive moats; AI erodes them at the application layer while deepening them at the infrastructure layer.
Marx predicted capitalism would produce the conditions for its own transcendence. He just didn't predict the mechanism would be a language model.
I don't know what comes after this. Nobody does. But I know it's coming faster than any of the frameworks we have can process.
The text box is free. The $690 billion is not. And both are building the same future — one we can type into but can't yet name.