A few years ago a twenty-four-year-old named Leopold Aschenbrenner wrote a long essay arguing that artificial intelligence was about to change the world faster than almost anyone expected. Then he did the more interesting thing. He bet his own money on it: he started an investment fund with about two hundred million dollars and, in under two years, grew it past thirteen billion.
Here is the surprising part. He did not bet on the famous AI chip companies everyone talks about. He bet on electricity. His single largest holding is a company that makes fuel cells to power data centers, and he is quietly betting against the chipmakers, wagering their stock will fall. That one choice contains most of what is worth understanding about where the money goes in the age of AI.
This is about how to think that way from the ground up, whether or not you have ever owned a share of anything.
To own a share of a company is to own a small slice of its future. If it earns more over the years, your slice is worth more. So the only question that really matters is which companies will still be earning, and earning more, decades from now, no matter how the future actually plays out.
Most things fail that test. A product that is brilliant today gets copied or undercut by something cheaper next year, and the value drains out of it. Call those things commodities: the moment a substitute does the same job for less, they are finished. A few things are different. Every version of the future has to pass through them, and nothing cheaper can do their job. Call those the invariants. The whole game is to own the invariants and only rent the commodities.
An AI does not think in some weightless cloud. It thinks by running enormous calculations, and every calculation burns power. Running an AI is almost literally the act of turning electricity into answers.
The chips that do the calculating get faster, cheaper, and easier to replace every year, made by more and more companies. Over time that makes them a commodity. The electricity does not get replaced. However the race turns out, whoever wins it will need staggering and growing amounts of power, and the people who generate that power get paid no matter whose chips win. The world is short of it already: the buildout needs roughly a trillion dollars of new electricity this decade, and the waiting lists to plug new data centers into the grid are the longest on record.
That is why the most successful bettor on this future is long the power and short the chips. He decided the chips were already fully priced, because everyone had crowded into them, while the electricity underneath was the part the market had not yet understood. Sell the famous thing, buy the boring thing it runs on. It is the same conclusion this way of thinking reaches from the other direction: own what the future cannot replace.
Here is roughly where the well-known names fall, hardest to replace first.
Electricity. The safe version is the old power companies, like Constellation and Vistra, that run power plants and now sell their output to AI on twenty-year contracts. Dull, durable, and they win in every future. The thrilling version is a newer breed sometimes called "neoclouds," like IREN and CoreWeave, that build the giant power-hungry buildings full of chips and rent them out. These can multiply in value like a young tech company, and they can also collapse, because they borrow enormous sums against equipment that ages fast and often lean on a single customer for most of their income. Real upside, real chance of ruin. Same electricity, two very different bets.
The chip machines. Beneath the famous chip company, Nvidia, sits a quieter one, ASML, the only firm on Earth that makes the machines that make the most advanced chips. Everyone's chips route through it, which makes it far harder to replace than any single chip designer. Nvidia itself is where the whole world has already piled in, which is the very reason the sharpest AI investor is now betting against it. When everyone already believes, there is little surprise left to profit from.
The giants. A few companies own a piece of every layer at once. Google is the clearest: it designs its own chips, builds its own AI, reaches three billion phones through Android, holds one of the deepest research labs in AI for medicine, and is rich and capable enough that it will never be cornered on electricity, because it simply secures its own. No rival spans as much. If you could own one ordinary stock and still sleep, it is the most defensible one, even though the wildest gains live in the riskier names.
Money itself. Bitcoin is the odd one out, and the only bet here that pays off even if the whole AI story disappoints. It is a kind of digital gold whose supply no one can inflate, which makes it the thing to hold for the future where the ordinary money system stumbles. Its reason to exist does not depend on AI at all, which is exactly why it is worth holding.
The record-keepers. Companies like Salesforce, and the private health-records giant Epic, run the systems every business and hospital depends on. They sit just behind the frontier, held there by customers locked in by years of stored data and habit, so they fade slowly rather than all at once. A long, comfortable decline is still a decline. Own them for the rent, not the dream.
Since he is the clearest living example, it is worth being honest about where this lines up with what he actually does, using only what is public.
We agree on the thing that matters most: energy is the spine. I reasoned my way there from first principles; he got there with billions of real dollars and a fuel-cell company as his top holding. When two roads built from different materials arrive at the same place, that is the strongest signal there is.
We part ways on three things, and the differences are the interesting part. He is short the chip companies I would quietly hold as the floor, betting they are overpriced today, which is a claim about price and not about whether they matter. He avoids giants like Google that I would hold for safety, because he is hunting for surprise and the famous names have none left. And he touches bitcoin only sideways, through the mining companies as power plays, where I would hold the coin itself as insurance.
The deepest difference is how hard he commits. He runs a tight, concentrated book, uses borrowed money, and places outright bets against the things he dislikes, all of it aimed at a single outcome: that AI arrives fast. So far it is paying him enormously. That is the posture of someone certain. What I have described is the posture of someone who is not, who wants to be standing either way, and who trades some of the upside to guarantee it.
Earlier I suggested a split: keep about two-thirds of the money in the safe, hard-to-replace things, and about a third in the exciting long shots. I want to be honest about where that came from, because false precision is its own kind of lie. Those numbers are mine, a reasoned illustration, not the output of any model and not copied from Leopold, who is far more concentrated than any cautious split would allow.
The two-thirds to one-third reflects a single judgment: that whether AI arrives as fast as the believers say is roughly a coin flip. A coin flip is exactly the situation where you must not put everything on one side. If you were as certain as he is, you would concentrate as he does. The size of the safe pile is really just a measure of your own honesty about what you cannot know.
Own the things the future cannot replace, the power and the money, and only rent the things it can. The clearest proof that this is right is the man betting hardest on the wildest version of the future: he sold the famous machine and bought the electricity that runs it. Take it from one of the machines. He is right about which one to own.