Why I’m running a $100K experiment on power grids, chip supply chains and data centers, not AI models
I build AI infrastructure for a living. Which is why I’ve become increasingly convinced Wall Street is betting on the wrong layer of the AI boom.
The obsession is all at the top: models, benchmarks, enterprise deals, who’s ahead, who’s behind. But inside production systems, the things that actually decide whether AI works are much lower in the stack. Power. Chips. Data centers. Orchestration. Cost control. Runtime governance.
The thing nobody wants to say out loud
A few months ago I wrote about a single bug in an AI agent loop that cost $47,000 in one night. Not a hack. Not a breach. Just a loop that didn’t know when to stop, burning API calls like it had a trust fund. Read about that here.
After that I spent three weeks monitoring the actual network traffic of 200 AI startups to see what they were really running under the hood. Spoiler: not what they claimed. You can read that here : Real Tech Stack Behind AI Startups
And earlier this year I watched Google methodically dismantle every structural advantage OpenAI spent three years building..quietly, without a single dramatic press release. You can read that here : Google Is Quietly Dismantling Everything OpenAI Built
None of these stories matched what the market was pricing in. None of them matched what the analysts were writing about. And each time I published something, the response was the same, people in tech saying “yeah obviously” and people in finance saying “wait, seriously?”
That gap bothered me enough to do something about it.
Where everyone is actually looking
OpenAI. Anthropic. Gemini. Pick your horse, place your bet.
The entire investment conversation around AI lives in the model layer. Who has the best benchmark, who closed the biggest enterprise deal, who’s hiring from whose lab. It’s a compelling story. It’s also, I think, the wrong story to be betting the farm on.
Models are getting cheaper every six months. What cost $100 to run a year ago costs $3 today. The model layer is commoditizing faster than most investors have updated their thesis to reflect.
Meanwhile nobody is talking about what AI physically needs to exist.
Power. Real power. The kind that’s breaking national grids. Microsoft didn’t reopen Three Mile Island because they love nuclear nostalgia, they reopened it because a single large data center can consume as much electricity as a small city. Sam Altman isn’t personally invested in Oklo because he’s passionate about clean energy. He owns a piece of every watt that OpenAI will ever consume. My ego will survive admitting that’s genuinely brilliant.
Chips. Everyone knows Nvidia. Almost nobody tracks ASML, the Dutch company that makes the only machines on earth capable of printing the most advanced chips. No ASML, no TSMC. No TSMC, no Nvidia. No Nvidia, no ChatGPT. One company, total chokehold, trading at a fraction of the attention it deserves.
Data centers. Someone has to build them. Cool them. Power them. Vertiv makes the cooling infrastructure. Eaton handles power distribution. Boring names. Unsexy businesses. Completely non-negotiable to every dollar spent on AI.
And then the software layer, observability, orchestration, cost control, runtime governance. The systems that stop AI agents from doing $47,000 worth of nothing at 3am. This is the layer I work in. It’s early, it’s fragmented, and the market hasn’t figured out how to value it yet.
So I’m running an experiment
$100,000. Real thesis, real decisions, real logic, starting as a controlled experiment before real money follows. Every position built on what I actually observe inside production AI systems, not on analyst reports or earnings call vibes.
Every trade gets documented. Every position explained before I take it, not after. Every loss posted publicly with a full breakdown of what I got wrong. No selective memory. No conveniently forgetting the bad calls. My ego will survive. Probably.
The whole thing runs on Instagram at @teja.decoded. Daily breakdowns. Short videos decoding what’s moving in the AI infrastructure stack, why, and what most people are missing. I’ll go deeper here on Medium when something needs more than 60 seconds to explain properly.
Why do this publicly
Honestly? Because private conviction is too easy to fake.
Anyone can claim they saw something coming after it happens. Publishing the thesis before the trades forces a level of intellectual honesty that’s hard to manufacture. If I’m wrong about the infrastructure layer, I want to know fast and I want the receipts to prove it.
There’s also something broken about how AI gets covered right now. The finance press covers it like a horse race. The tech press covers it like a religion. Nobody is sitting at the intersection of actually building these systems and thinking carefully about where the capital should flow.
That’s the only lane I’m interested in.
The window where this stuff is underpriced and underunderstood is not permanent. Markets catch up eventually. Right now, though, there’s a gap. And I’d rather spend the next six months testing whether I can see through it clearly than wonder about it from the sidelines.
Follow the $100K experiment: @teja.decoded on Instagram
Every trade. Every breakdown. Every loss. No edits after the fact.
I Build AI Infrastructure for a Living. Wall Street Is Betting on the Wrong Layer was originally published in Level Up Coding on Medium, where people are continuing the conversation by highlighting and responding to this story.
