I Stopped Predicting AI. Here's What I Do Instead.

Everyone has an AI prediction.

Some are societal: utopia or extinction. AI ends disease, poverty, boring work. Or AI ends the jobs, the meaning, the point of what most people do. Everything humans thought made them irreplaceable.

Some are about the market: which company wins, which valuation is insane, which model survives. Confident takes on Nvidia's P/E. Strong opinions on OpenAI's next move.

I had those too. Said them out loud. Was wrong.

So I stopped.

I stopped predicting. I started building instead.

A structure. Built from five years of watching, reading, rethinking. And one moment when it confirmed itself.

For eight years I ran an agency. You learn to feel the wind.

Around 2022, 2023, the wind turned. Against the agency model. Against every assumption we'd built on. AI had arrived. It was already rewriting what clients needed from us.

I started looking hard for where to go.

The first thing I saw: AI was moving faster than anything I'd tracked. Not incrementally. Vertically. Not affecting some industries. Affecting everything. Everyone.

The second: history had a pattern. Steam. Electricity. Internet. Each General Purpose Technology followed the same arc. Infrastructure first. Speculation next. Productivity wave after. Each cycle shorter and bigger than the last.

The third: power law. Technology value doesn't spread evenly. It concentrates. Look at the NASDAQ over twenty years. But look closer: even inside the NASDAQ, it was six or seven companies pulling almost everything. Mag7 wasn't a fluke. It was how technology waves work. With AI, the concentration gets more extreme. Frontier models cost billions to run. The winners will be fewer. They'll accumulate more.

I started moving capital. Not broad. Concentrated. I started with the NASDAQ: less drag than the broad market, no dead weight from the old economy. Then I looked inside. And I saw legacy still there. Companies from a different era, carried along by the index. Companies that won't ride the next wave. They'll survive it at best.

That bothered me. So I drilled further.

Semiconductor and AI infrastructure indices. The physical layer every model, every agent, every AI application runs through. Not a bet on one company. A bet on what every version of the winning scenario requires. If semiconductors are the infrastructure layer, I don't need to know which chip company leads. I need to be in the layer.

Then the data confirmed it.

Semiconductors compounded at approximately 18% per year for the past 19 years. The S&P 500 at 10%. The difference sounds like a rounding error. Over 19 years, it's roughly four times the outcome. I ran the full numbers elsewhere.

I'd been connecting dots for years. The confirmation had been sitting in the data the whole time.

I remember reading those numbers and feeling something settle. Not excitement. Something quieter. An argument I'd been making privately, suddenly supported by evidence I hadn't looked for yet. It was already there. It had already happened. And if the AI wave is bigger than the internet wave, as most economists who study this believe, then 18% is not the ceiling.

It's the floor.

Here's what I do.

Keep weight down. In life and in the portfolio: no legacy, no drag, nothing that can't compound.

Build optionality. No country claims me. No single asset traps me. No scenario closes me off permanently.

Invest in bottlenecks. Concentrated. Structural. The layer beneath every possible AI outcome, not the application layer with its short half-life of certainty. The full structural argument for why, and why not data centers or energy, is here.

When the thesis was clear, the conclusion was obvious. I sold everything.

Not in panic. Over a few weeks, position by position. Each sale made the next one easier. I still believed in the companies. But believing in a company and owning its stock aren't the same thing. I wasn't built to track earnings. To read quarterly reports. To know when to cut. That's not my game.

By the end, something had settled. Not just in the portfolio. This is a strategy I can follow for twenty years. I don't need to check it every quarter. I just need to stay in it.

Here's what this changed for me.

Adobe fell over 60% in the last two years. The fundamentals barely moved. Gross margins above 85%, recurring revenue, a product I have used since university. I had a target price I'd written down years earlier. It hit. Then went below it. The moment I'd been theoretically waiting for was right there.

I didn't buy. It doesn't fit the thesis. Adobe is a software layer that AI is repricing. Brilliant product, wrong position for where I want to be. No single stocks. No stock picking. The infrastructure layer only.

I stopped watching AI content. When something catches my attention, I pull the transcript and run it against my thesis using an LLM. Counter-argument or confirmation. Either way, I update or I hold. I don't sit through the video. I don't follow the creator. I don't need to be informed. I need to be positioned.

That's a weight off you don't expect until you put it down.

Intelligently positioning yourself is the most optimistic thing I can think of in an uncertain world. Not because you've predicted correctly. Because you've built something that doesn't require a prediction at all.

So I stopped predicting.

Not advice. Five years of dots connecting — and a structure that works wherever the wave breaks.