The market panicked over one word last week

On July 1, a lot of AI-related stocks dropped hard, fast, and mostly for the wrong reason. The trigger was a Bloomberg report that Meta ($META), the company behind Facebook and Instagram, is planning to rent out some of its artificial-intelligence computing power to other companies. Meta's own stock jumped almost 9% on the news. Then a strange thing happened. Nearly everything else connected to AI fell at the same time, and some of it fell a lot.

The story the market told itself in that moment went like this: if Meta has spare computing power to rent out, then the AI companies must have bought too much hardware, the boom must be cooling, and the whole build-out is about to slow down. On that single thread, chipmakers sold off, memory makers sold off, and a group of smaller AI-computing companies sold off even harder.

Most of that was a misread, and it is the kind of misread that shows up often enough to be worth learning to spot. So this issue does two things. Above, the simple way to see what actually happened, using a picture you already understand. Below, the specific answer to the question that matters: who genuinely has a problem here, and who just got dragged down for nothing.

My first book, Make Your Own Alpha, is on its way. It is the full version of the approach behind these issues: how to build a portfolio deliberately enough to make work optional, without a finance background or a windfall to start.

The book is launching this Thursday July 9. Click “Yes, I’m in!” to get on the notification list for the special launch bundle and pricing.

Start with what Meta announced, because the plain-English version is far less dramatic than the stock move.

Meta is spending an enormous amount of money, somewhere in the range of $125 to $145 billion this year by its own guidance, building AI computing capacity for its own future needs. That build-out runs years ahead of what the company can use today, on purpose, because it expects to grow into it. In the meantime, a chunk of that capacity sits available. So Meta decided to rent some of it out to other companies while it grows, to earn money back on an investment it was going to make anyway.

That is the whole event. A company that is building for the long term found a way to make its shorter-term spare capacity pay for itself.

Imagine a company builds a hundred-story apartment tower because it can see, years ahead, that it is going to need every floor. Today it needs maybe forty of those floors. So rather than let sixty floors sit empty while it grows, it rents them out to tenants and uses the rent to help cover the mortgage. Smart, ordinary, and not remotely a sign of trouble.

Now imagine the stock market watches this landlord list some empty units for rent and concludes, "There must be too many apartments in this city. The building boom is over. Sell every construction company in town." That is close to what happened last week. Meta listing its spare capacity got read as proof that the world suddenly has too much computing power, and everything tied to building that computing power fell together.

What we already know about demand is the tell that this was a misread. Meta is not a company drowning in spare computing power. In its most recent quarter it committed something like a hundred billion dollars in new infrastructure contracts, and reporting this year described Meta being told by another tech giant that it could not buy all the capacity it wanted, because that supplier was short too. When the biggest players are still being turned away at the counter, "too much supply" is the wrong description. The right one is closer to a shortage that is being carefully rationed.

There is even a quiet clue in Meta's own decision. A company does not open a rental business in a market where renting loses money. By choosing to rent its capacity out, Meta is effectively telling us the rents are high enough to be worth collecting. That points in the opposite direction from "the boom is ending."

Who rents out the computing power is a landlord question. How much hardware it took to build that computing power is a construction question. They are not the same question, and last week the market answered the second with the first.

Go back to the apartment tower. When the landlord rents out the empty floors, does that change how much steel, concrete, glass, and wiring it took to put the building up? Of course not. The building materials were already bought and installed. All that changed is whose name is on the leases. The construction bill is identical whether the landlord fills those floors with its own staff or with paying tenants.

Computing power works the same way. The chips, the memory, and the networking gear that make up Meta's spare capacity are already bought and already installed. Whether Meta uses that capacity itself or rents it to someone else changes who is using it, not how many chips it took to build it. So when the chipmakers and memory makers fell last week on the idea that Meta renting out capacity means the world needs less hardware, the market was charging the construction companies for a change in the tenant list.

That is the whole misunderstanding in one sentence: the market confused landlord drama with construction demand.

Hold that rule, because it does most of the sorting. Most of what fell last week had nothing to do with the news, and a narrow slice genuinely did. Separating the two is the entire job, and it is the rest of this issue.

The largest group of victims last week had almost nothing to do with the announcement: the companies that make the hardware. The chipmakers, the memory makers, the networking-chip makers. Names like NVIDIA ($NVDA) and Micron ($MU) fell alongside everything else, on the idea that Meta renting out capacity means the world needs fewer chips.

Go back to the tower one more time. When the landlord rents its empty floors to tenants instead of using them itself, not one brick changes. The building is already built. The chips inside Meta's spare capacity are already bought and installed, so whether Meta runs that capacity itself or rents it out changes who is using it, not how many chips it took to build it.

The evidence that hardware is still scarce, not piled up, is not subtle either. Meta itself was being turned away by a supplier earlier this year. The giant cloud companies that sell computing power are so short that at least one of them has been renting emergency capacity from others at close to a billion dollars a month. When both the buyers and the sellers of computing power are short at the same time, that is a shortage being rationed, not a glut being dumped. So the layer that fell hardest last week, the chipmakers, was the one with the least to do with the news.

The companies that build and assemble the physical clusters win no matter which landlord ends up owning the cloud. These are the picks-and-shovels names, the ones that supply every side of the fight.

Celestica ($CLS) has been posting record margins as its work shifts from simple assembly toward building whole racks and clusters, and it sits squarely on the benefit side, because more companies building computing power, Meta very much included, means more orders for the company that assembles it. The large contract manufacturers behind the scenes, names like Quanta and Foxconn, ride the same wave. And Penguin Solutions ($PENG), which I hold, runs large computing clusters as a service, which makes it genuinely neutral to who wins the rental business, since somebody has to build and operate the clusters either way.

More competition to rent out computing power means more buildings going up, which is more work for the builders and more chips for the suppliers, not less. That is the quiet punchline the headline buried.

The one group with a genuine question to answer is the set of smaller companies whose whole business is being the landlord, the ones now called neoclouds. They borrow heavily to buy chips and rent them out, so a giant like Meta stepping onto their turf is a real change. It works through their loans and their future contracts rather than through a price war tomorrow, since through at least 2027 Meta stays a net buyer of computing power. The question is whether these companies' contracts still get renewed years from now, once Meta both supplies itself and sells to others. That is a question for their lenders today, not a hit to this year's revenue.

The most interesting split, though, sits inside that group. Two names that fell almost identically last week, CoreWeave ($CRWV) and Nebius ($NBIS), are facing opposite realities. One has a real problem to work through. The other was sold on a fear that its own contract with Meta directly contradicts. That contrast is where the biggest opportunity and the biggest risk in this whole story sit, so it is getting its own dedicated issue next, where I put the two side by side and show why the difference matters more than the shared headline.

Real investing is mostly the patient work of looking past a frightening headline to ask a calmer question: who here actually loses revenue, and who just had their stock shoved around by a crowd reacting to a word? Last week the word was "surplus." It pulled the chipmakers, the hardware builders, and a set of very different landlords into one drop, whether or not the word had anything to do with them.

The builders and the chipmakers had almost no business falling. One kind of landlord did. Telling those apart, calmly, days after the crowd has moved on, is a large part of what building a portfolio you can rely on comes down to.

Stay disciplined - Koh

Disclaimer: Nothing in this newsletter constitutes investment advice or a recommendation to buy or sell any security. Numbers and observations are as of publication. I may hold positions in companies discussed above. Always do your own research and consult a licensed financial advisor before making investment decisions.