The market spent June selling the companies that buy AI, and buying the ones that sell it

The number everyone repeated as June closed was $2.3 trillion — the market value that came off the Magnificent Seven over the month, a figure with enough zeros to function as a headline on its own. Microsoft fell around 20%, Nvidia roughly 13%, Apple and Amazon about 8% apiece. As a group the seven were down about 3.4% on the month, which sounds mild until you remember that 3.4% of these companies is most of a Norway. The word attached to it, again and again, was "jitters": investors growing nervous about how much these companies are spending on artificial intelligence and how little they can yet show for it.

That is the headline. The more interesting number sat one line down, and almost no one set the two side by side. While the companies that buy AI infrastructure were selling off, the companies that sell it were doing the opposite. The Philadelphia Semiconductor Index — the chipmakers, the memory houses, the equipment makers — rose about 6% in June and is up more than 90% on the year. Micron, on the strength of a record quarter, briefly passed Meta in market value. So the month did not deliver a verdict on the AI trade. It delivered a rotation inside it: out of the buyers, into the sellers. The question worth asking is whether that rotation is the prudent move it looks like, or the same position with a different label on the risk.

What actually moved, and what didn't

Start with the mechanism, because "jitters" is a mood, not an explanation. The capital expenditure tied to AI is on track to top roughly $700 billion across the hyperscalers in 2026, up from something like $387 billion the year before — call it a 70-to-80% jump, depending on whose tally you use, in a single year. That money goes out the door now, as chips, land, transformers, and construction. The revenue that is supposed to justify it arrives later, in increments, and so far in amounts that do not obviously close the gap. For a company like Microsoft or Amazon, that spending lands on the financial statements as depreciation and as a heavy drag on free cash flow long before it lands as profit. The market spent June marking down the present value of that bet — not deciding it was wrong, but deciding it would no longer pay a premium for the privilege of waiting to find out.

Now follow the same dollar to the other side of the trade. Every one of those capex dollars is somebody's revenue. The $700 billion the buyers are spending is, in large part, the order book the sellers are being repriced on. Nvidia's data-center revenue, Micron's record quarter, the equipment makers' backlogs — these are not an independent good-news story running parallel to the hyperscaler build-out. They are the build-out, viewed from the receiving end. So the market in June did something quietly strange: it grew more skeptical of the spending and more enthusiastic about the spending's beneficiaries, at the same time, in the same breath. That is not a hedge. It is a bet that the buyers will keep writing the cheques even as their own shareholders punish them for it.

Selling the buyers and buying the sellers is not a hedge. It's a bet that the cheques keep clearing after the people writing them have been told to stop.

Why the rotation feels like caution

There is a respectable logic to preferring the sellers, and it is worth stating fairly. In a gold rush, the reliable money is in shovels — the supplier gets paid whether or not the miner strikes anything. A chipmaker books revenue the moment the order ships; it does not have to prove that the customer's AI product found a market. The semiconductor names also have something the hyperscalers currently lack: visible scarcity. Memory is genuinely tight, advanced packaging is genuinely constrained, and pricing has moved accordingly. When a thing is scarce and selling at a premium, the income statement is easy to believe. When a thing is a $700 billion promise about future demand, the income statement is an act of faith. Investors in June chose the easy belief over the act of faith, and on a one-month view it worked.

Notice, too, where the money went when it left the seven. It did not leave the AI theme; it moved within it, and it moved toward the part of the chain with the clearest near-term numbers. That is what a rotation is — not capitulation, but a reallocation by people who still want the exposure and have simply lost their appetite for the version of it that asks them to pre-pay for results. Plenty of the selling was also ordinary book-squaring at a quarter-end, in names that had run a very long way. None of this requires anyone to believe the boom is over. It only requires them to believe that the toll-collector is a safer seat than the road-builder. For now, that belief is correct.

The exposure the rotation doesn't remove

Here is where the trader's reflex kicks in, the one that asks what breaks if the story is wrong. The comfort of owning the sellers rests on an assumption that is rarely said out loud: that the buyers' spending is durable, that the capex line only goes up or, at worst, plateaus. The entire case for the chipmakers' backlog is the permanence of the hyperscalers' build-out. But the thing the market spent June doing was questioning exactly that permanence. You cannot, coherently, hold both views at full conviction — that the hyperscalers are over-spending on AI, and that the companies whose revenue is that spending are a safe harbour from the doubt. The safety is borrowed from the very confidence being withdrawn next door.

This is the concentration risk hiding in plain sight, and it is structural rather than dramatic. The supplier rally and the buyer build-out are not two bets; they are one bet, observed from two ends of the same invoice. A handful of customers — four or five hyperscalers — account for an outsized share of the demand under the entire semiconductor complex. That is concentration of the most dangerous kind, because it does not look like concentration. It looks like diversification: you sold the megacaps and bought the chip names, you feel rotated, you feel hedged. You are not. If the capex plans that the market started doubting in June are actually trimmed — even modestly, even at the margin — the revisions do not stay with the buyers. They travel straight down the invoice to the sellers everyone just crowded into, and they arrive as a cut to the one number, the order book, that made the sellers feel safe in the first place.

The signs the spending might bend

None of this is a forecast that the capex will fall. It is an argument about where the exposure sits if it does, and there were enough small signals in June to make the question live rather than academic. A few are worth keeping an eye on:

• The financing is getting more visible. The build-out has increasingly been funded with debt and off-balance-sheet structures rather than out of operating cash flow — a sign that even cash-rich companies are starting to feel the cost of carrying it, and a layer that behaves badly if sentiment turns.
• The buyers are being rewarded for restraint, not ambition. June's price action punished the heaviest spenders most; Microsoft, the most aggressive, fell the hardest. When the market starts paying companies to spend less, boards eventually listen.
• The data-center supply chain is showing its own stress. Digital Realty fell about 5% late in the month after taking a $3.5 billion stake in Blackstone-linked data-center assets — a reminder that the property-and-power layer of the boom carries its own leverage and its own skeptics.
• The returns case is still mostly promissory. The revenue that is meant to justify $700 billion of annual spending is real but early, and the gap between the spend and the payback is precisely what the second-quarter numbers will be read for.

Each of these is small on its own. Together they describe a market that has started, for the first time in this cycle, to treat the capex as a cost to be questioned rather than a virtue to be applauded. That shift in posture is the actual news of June, more than any single percentage.

What the July numbers have to settle

Which is why the second-quarter earnings season, beginning in July, is doing more work than usual. It is being asked to settle an argument the market could not settle on sentiment alone: whether the AI spending is generating a return fast enough to outrun the depreciation it creates. Dan Ives at Wedbush, among the build-out's more durable bulls, has framed the coming reports as the validation the thesis needs — and that framing cuts both ways. If validation is what July is for, then July is also when the thesis can fail to validate. The reports will not be read for revenue beats in the usual sense. They will be read for the ratio between what these companies are spending and what that spending is starting to earn, and for any language from the buyers that hints the capex curve might bend.

For the chip names that rallied through June, this is the dependency that does not show up in their own results. A semiconductor company can report a flawless quarter and still be repriced the next morning on a single sentence from a customer's call — a softened capex guide, a pushed-out data-center timeline, a note about "optimizing" spend. The seller's stock is only as good as the buyer's conviction, and the buyer's conviction is exactly what spent June eroding. That is the basis-point reality behind the rotation: the people who sold the Magnificent Seven and bought the shovel-makers did not reduce their exposure to the AI capex cycle. They concentrated it, and then relabelled it as prudence.

Separate the companies from the trade, the way you always should. Nvidia, Micron, TSMC, ASML — these are extraordinary businesses, and nothing here says otherwise. But a great company and a safe position are different claims, and the rotation of June quietly bundled them back together at exactly the moment the market was, next door, pulling them apart. The downside no one is pricing is not that the suppliers are bad. It is that they are the same bet as the buyers, made one invoice later, by people who think they have moved to higher ground. If the capex holds, they have. If it bends, they will discover that the road-builder and the toll-collector were always standing on the same road.