A junior developer in Austin wires up another feature. The task is to add an AI-powered summary to a new client’s customer service dashboard. He doesn’t spin up a server or train a model. He writes a few lines of code to call an API endpoint at a company in San Francisco, wraps the output in a clean UI, and pushes the update before lunch. All across the world, thousands of developers are doing the same thing. They are building a new world on borrowed intelligence.
This is celebrated as a democratization of technology. It is not. It is the construction of a monoculture, a glittering, complex ecosystem built atop a handful of foundations so vast and expensive that only a few corporations can ever hope to build them. The Cambrian explosion of AI applications we see today—the clever writing assistants, the coding copilots, the marketing slogan generators—is a facade. Beneath it, the digital economy is quietly developing a critical dependency, a single point of failure that is planetary in scale.
Look past the logos on the latest AI-powered app. You will likely find the same engine humming underneath. A startup promising to revolutionize legal discovery and another vowing to disrupt graphic design are often just sophisticated prompt engineers renting cognition from the same source: OpenAI’s GPT-4, Anthropic’s Claude 3, or Google’s Gemini. The innovation isn't in creating intelligence, but in finding clever ways to package and resell it. This creates a market of tenants, not owners. These companies are building their businesses in a skyscraper they not only don't own, but whose blueprints they are not allowed to see.
The risk here is not merely economic. It’s systemic. When a single strain of wheat dominates agriculture, a single blight can lead to famine. We are creating the same vulnerability in our cognitive infrastructure. A subtle political bias, a security flaw, or a persistent inaccuracy in a foundational model is not an isolated bug. It is a poison that seeps into thousands of downstream applications simultaneously. One day that flaw might manifest as biased hiring recommendations across an entire industry. The next, it could be a persistent error in medical diagnostic tools used by hospitals that have all licensed software from different vendors, never realizing they share the same flawed digital brain.
The companies that control these foundational models—Microsoft, Google, Anthropic, a few others—are not just service providers. They are becoming central banks for thought. They set the price for reasoning. They can alter the capabilities of their models with a software update, obsoleting entire categories of startups overnight. They hold the keys to the intellectual infrastructure of the 21st century, and their boardrooms are the new chokepoints for global commerce and information.
The frenzy in Silicon Valley is focused on the clever applications, the next thin wrapper that can find a market niche. But the real power is being consolidated at the base of the stack. While venture capitalists fund an army of tenants decorating their apartments, the giants are fortifying the foundation and installing the meters. The next great tech crisis won't be a market crash or a server outage. It will be a crack in one of these brittle, monolithic foundations, and we will all feel the tremor.