When X-rays were first discovered, they weren’t immediately used to set broken bones or treat cancer. For a while, they were a carnival attraction. People paid to see the skeleton inside their own hand as a curiosity. It took decades for the technology to mature from a circus trick into the backbone of modern medicine.
I feel like we are in the “circus” phase of Artificial Intelligence.
We are in Year 3 of the Generative AI revolution. To compare this to electricity, we haven’t even invented the lightbulb yet; we are still marveling that we can make a spark jump between two wires. We have had hundreds of years to optimize our world around electricity. With AI, we are just getting started.
There is a lot of talk about an “AI Bubble.” As someone who spends every day building AI applications, I have a front-row seat. My take? The financial bubble might be real, but the product future is brighter than anyone realizes.
The Return of the Cowboys
For a product designer, this is the most exciting time in history.
For the last decade, tech has been a mature industry. It was about “ruthless execution.” Every idea had been done; the winner was just whoever could optimize their funnel the best.
AI has brought back the “Early Internet Cowboys.” We don’t know what the ultimate form factor of this technology is yet. We are making it up as we go.
The future isn’t just “chatbots.” The best AI will be invisible. It will be background processes, thousands of API calls a day that you never see, making your life smoother, less annoying, and more independent. We are moving away from “Look at this AI!” to “Look how easy this task became.”
The “Moat” Illusion
However, while I am bullish on using AI, I get shivers when I look at the valuations of the companies making the models.
I recently switched from OpenAI to Google’s Gemini. I had been a paid subscriber to ChatGPT since GPT-3.5. It was synonymous with AI for me. But recently, they felt like they were falling behind.
The switch was terrifyingly easy.
I took my client files—complete notes of every meeting—started a new thread in Gemini, uploaded the data, and boom. Gemini was trained. In seconds, it was performing better than my old ChatGPT setup.
There was zero friction. There was no “moat.”
If I can switch my entire business logic from one AI provider to another in the time it takes to drink a coffee, where is the defensibility for these trillion-dollar model builders?
Coke, Pepsi, and DeepSeek
This brings us to the “Coke vs. Pepsi” dynamic.
The big model providers (OpenAI, Google, Anthropic) are fighting to be the preferred brand of “intelligence tokens.” But then came DeepSeek. They proved that you can get “good enough” AI for a fraction of the cost by being a fast follower.
If “good enough” AI becomes extremely cheap—or if we can all run decent models on our own private servers—the margins for the big players will collapse. They are laying the fiber optic cables of the 21st century, just like the telecom boom of the late 90s. The infrastructure is vital, but the infrastructure builders might not be the ones who make the profit.
The Real Winners
So, if the model builders are just selling electricity, who wins?
The appliance makers. The people who figure out what to do with the electricity.
I suspect the AI operators themselves will make the least amount of money in the long run relative to their investment. The real value is in the application layer.
It takes so little now to make a great app. A single person or a very small team can build software that rivals what used to take a hundred engineers.
That is the category of company I want to see. Not the ones building the power plant, but the ones inventing the lightbulb, the toaster, and the X-ray machine.
That is exactly what we are doing at InfiniteUp. The bubble may pop for the infrastructure, but for the builders, we are just getting started.
Optional Title Ideas
From Circus Tricks to Medicine: Why We Are Still Early in AI
Why I Switched to Gemini (And Why It Should Scare OpenAI Investors)
The AI Moat is a Myth: A Developer’s Perspective
Cowboys, Utilities, and the AI Bubble
Social Media Snippets (for LinkedIn/Twitter)
Option 1: I switched from OpenAI to Gemini in 5 minutes. No friction. No switching costs. If the “moat” is that shallow, what does that say about the AI bubble?
Option 2: X-rays started in circuses before they revolutionized medicine. We are currently in the “circus phase” of AI. The real utility is coming, and it won’t look like a chatbot.
Option 3: The big AI labs are fighting a Coke vs. Pepsi war. Meanwhile, small teams are using their cheap “sugar water” (tokens) to build the next generation of world-changing apps.
Questions for you to refine the piece:
The “Shivers”: You mentioned the easy switch to Gemini. Do you want to elaborate on why Gemini felt better? Was it the context window, the reasoning, or just the freshness of the model?
InfiniteUp: Do you want to hint at what InfiniteUp is building specifically, or keep it vague as “the app that changes the world”?
Invisible AI: Do you have a specific example of “Invisible AI” you’d like to include? (e.g., “Your calendar negotiating with your dentist’s calendar without you ever opening an email”).