The chip trade and the hyperscaler trade just went their separate ways

Chip stocks are having one of their best stretches in years.

The companies buying those chips are having one of their worst.

That split shouldn’t happen if the AI story is as unified as it looks on the surface. Chipmakers sell to hyperscalers.

Hyperscalers selll AI services to everyone else.

If one leg of that chain is struggling, the other leg usually feels it too, eventually.

Instead, memory names have been on a tear….

Micron’s market cap crossed a trillion dollars.

Meanwhile the Magnificent Seven, the group whose capex has been funding this entire buildout, dropped more than 12% in June alone.

Two narratives showed up in my feed within days of each other. One says this divergence is nothing to worry about.

The other says it’s the exact pattern that showed up before the dot-com crash.

Both narratives can’t be right at the same time, and neither one is obviously wrong either.

This is the kind of split where runing the story through a structured process beats picking a side because it feels right.

JPMorgan flagged the pattern directly. A team at the bank pointed out that a growing gap between big AI spenders and hardware stocks also showed up in the months before the dot-com bubble burst, and warned that the same setup could eventually pressure the semiconductor trade too. That’s not a throwaway line. It’s a bank naming a specific historical parallel and putting a number on the risk.

At the same time, the same JPMorgan note said the bank was still leaning bullish overall. That’s the tension in one sentence. The people flagging the risk aren’t actually calling the top.

On the other side, some analysts describe the rotation as neither bullish nor bearish, just capital moving toward whichever part of the AI trade is showing the clearest results right now, which currently happens to be memory chips.

So you’ve got a real historical parallel on one side and a shrug on the other. That gap is worth taking seriously before you decide which camp you’re in.

When a story splits into a scary parallel and a shrug this cleanly, I run it through three questions before I decide what to do with it. Here’s the version I used on the chips versus hyperscalers split. Copy this into Claude the next time you see a rotation that has the market talking past itself.

Here is a market narrative comparing a current rotation in [sector or trade] to a specific historical period.

Part 1: What is actually being compared State the historical period being referenced and the specific metric or pattern that is said to resemble today’s market. Keep this to the mechanical comparison only, not the conclusion drawn from it.

Part 2: Where the comparison holds List the specific data points where today’s situation genuinely resembles the historical period cited. For each one, name the source, whether that’s a bank note, a company filing, or market data, and how it was measured.

Part 3: Where the comparison breaks down List the meaningful differences between today’s situation and the historical period. Focus on structural differences, such as company profitability, balance sheet strength, or revenue quality, not just sentiment.

Part 4: Who is making the comparison and what they actually concluded Name the source of the comparison and state their actual stated view, including any hedges or caveats they included. Do not let a scary headline stand in for what the source actually said if their own conclusion was more measured.

Keep the language plain. If a claim can’t be traced to a named source, say so directly instead of softening it.

This step alone stopped me from overreacting. The dot-com comparison came from a real JPMorgan note, but the same note said the bank still leaned bullish. The headline version of that story loses the second half entirely.

Now take the bearish narrative around this rotation specifically.

Part 1: State the bearish case in one sentence

Part 2: What has to be true List the specific financial conditions that would need to hold for the bearish case to play out. Be concrete. Instead of “capex needs to fall,” say whose capex, over what timeframe, and roughly what percentage decline would count as confirming.

Part 3: What’s already on the record Look at the most recent earnings calls and guidance from the companies at the center of this trade. Tell me if anything already disclosed leans toward the bearish case, or if the evidence genuinely isn’t there yet.

Part 4: The one number that decides it Tell me the single data point, such as a specific capex guidance revision or a segment growth number, that would most clearly confirm or kill the bearish case if it showed up in the next earnings cycle.

Part 5: Timing Name the specific event, such as which company’s earnings call and roughly when, that would produce that number.

Running this on the current setup pointed to one clear checkpoint. The bearish case needs hyperscaler capex guidance to come down in the next earnings cycle. Until that happens, the dot-com comparison is a pattern, not a confirmed repeat.