Oct 21, 2025 · 5 min read
Tech giants are pouring hundreds of billions of dollars into chips, data centers, and power plants, creating a financial phenomenon that rivals the biggest industrial booms of the past. The scale of this spending is hard to comprehend, comparable to the capital expenditure of the entire global oil and gas industry.
This massive surge in spending feels like a bubble, and history tells us it probably is one. From railways to the internet, every major technological shift triggers a period of frenzy where excitement outpaces reality, and investors assume growth will never stop. But the existence of a bubble does not mean the technology isn't real. The internet was a bubble in 1999, yet it completely rewrote how the world works. The dust will eventually settle, and we will be left with a new global infrastructure that changes everything.
While the infrastructure grows, companies like OpenAI, Google, and Anthropic are spending billions to train "frontier" models, but the gap between the best model and the second-best is shrinking.
A rapid convergence is happening that would have been unthinkable just a few years ago. Having the best model was once a massive advantage, but today the leaders change weekly, benchmarks are saturated, and for general tasks, the difference between the top competitors is often negligible.
This creates a difficult situation for the model labs, who are engaged in a massively expensive science project where the output is becoming a commodity. If everyone has a "super-brain," having one isn't a unique selling point anymore. It becomes table stakes, like having electricity or internet access.
Because the intelligence itself is no longer a guaranteed defense, model companies are frantically trying to build a "moat," a protective barrier around their business to keep competitors out.
OpenAI's strategy illustrates this clearly. They are saying "yes" to everything at once. They are not just a research lab anymore. They are making infrastructure deals with Oracle and buying chips, building consumer apps, web browsers, and search engines, integrating with e-commerce and exploring robotics. This is a race to turn a raw technology into a sticky product, bundling their model with enough other services and infrastructure that users, and investors, have a reason to stay, even if a cheaper or slightly better model appears tomorrow.
If the models in the middle are commodities, the value capture has to move elsewhere, and it tends to move in two opposing directions.
First, it moves down the stack to the physical assets, where the companies that own the scarcity, the chips, the power plants, and the massive data centers, have a guaranteed revenue stream. They are selling the pickaxes for the gold rush.
Second, it moves up the stack to the product layer, where if everyone has the same "brain," the winner is the company with the best workflow, the best user experience, and the deepest network effects. The advantage isn't the AI; it's the proprietary data the AI can access and the complex software that wraps around it.
While model companies fight for dominance, the users of this technology are benefiting from a new economic reality where the cost of intelligence is dropping toward zero.
This creates a dynamic known as the Jevons Paradox, where when you make a resource more efficient and cheaper, people don't use less of it; they use massive amounts more because they find new things to do with it. AI gives companies a supply of "infinite interns," and this doesn't just mean automating existing jobs. It means doing things that were previously too expensive to attempt. We might generate ten thousand variations of an ad campaign instead of three, or write custom software for a single-use project that is discarded a week later. We will expand the scope of what is possible rather than just shrinking the workforce.
The true disruption happens when we stop just adding AI to old software and start building new types of software. Right now, we are mostly in the "Absorb" phase, where we add chatbots to existing apps, but the future is the "Innovate" phase.
The question is what gets unbundled. The internet unbundled the newspaper and the music album, and AI might unbundle the very concept of an "app." If an AI can instantly generate the user interface you need for a specific task, do you need a complex dashboard with a thousand buttons?
The world is moving toward software that adapts to the user, rather than the user learning the software. The rigid structures of menus, tabs, and file systems are artifacts of a time when computers were dumb, and as computers get smart, the software should disappear, leaving only the solution to the user's problem.