How to classify your AI holdings: the reset / crack / hype test

Companion article to Issue 13 // 5 June 2026 // R. Lauritsen

How to classify your AI holdings: the reset / crack / hype test

This week the market punished a 200% grower and rewarded a one-sentence compliment. The newsletter showed you the regime shift; this is the tool to act on it — a five-step test that sorts any AI name into one of three boxes, the map that shows why, and the exact number that would reclassify each one.

9 min read // Sectors: semiconductors, AI software, robotics, macro

The setup, and how to spot it before earnings

Quick recap for anyone arriving cold: a stock moves on results relative to what was already priced in, not on whether they’re good in absolute terms. After a two-month melt-up that carried semis and AI software 30–60% higher, an enormous amount of good news was already in the price. Broadcom grew AI revenue 200%+ and fell 13% (it reiterated rather than raised); CrowdStrike beat, raised and split, and fell 9%. No bad numbers — just expectations the prints couldn’t clear. That’s the regime in two sentences.

Here’s the part the newsletter didn’t have room for: you can often see this setup coming. The whisper rises faster than consensus in three detectable conditions, and all three were flashing before this week’s prints:

• A steep run into the print. When a stock is up 40%+ in two months, the buyers setting the marginal price already expect a blowout. A merely-good number disappoints them by definition. (Rule of thumb: the bigger the pre-earnings run, the higher the invisible bar.)

• Sell-side targets clustering above consensus. When analysts start raising price targets in the fortnight before a report while leaving official estimates alone, the gap between the two is the whisper. Watch the target revisions, not the headline estimate.

• Options pricing a big move. An elevated implied move (what the options market expects the stock to swing on earnings) means traders already know a normal beat won’t cut it. A 10%+ implied move into a report is the market admitting the bar is sky-high.

None of these tells you which way a stock breaks. They tell you the bar is high enough that good news may not be rewarded — which is exactly when a beat-and-fall becomes likely, and exactly when the classification below earns its keep.

The three boxes every AI name falls into

Most commentary this week sorted stocks into “bubble popping” (it fell) or “boom continues” (it rose). That’s the lazy split. The useful one has three boxes, because a falling stock and a rising stock can each mean two very different things. Get the box right and the response is obvious; get it wrong and you either sell a great business into weakness or buy a story with nothing under it.

The reset and the crack look identical on a one-day chart and opposite on a one-year one — that’s the entire trap. The hype run is the one most likely to feel like genius and end like a hangover.

The five-step test: classify any name in two minutes

This is the tool. Run any AI stock that moved hard through these five questions, in order, and it lands in exactly one box. It works on names not mentioned in this issue — that’s the point.

Worked example — an ambiguous one. Micron fell 8% this week, but it didn’t report. Step 1–2: no print, no guide change. Step 3: no fresh number — it moved on Broadcom’s read-through. Step 4: the drop is neither its own valuation nor its own business; it’s sector contagion. That makes MU a contagion case — a reset-by-association, not a crack — and Step 5 names the resolver: the 24 June print. If HBM pricing holds, the 8% was someone else’s mood; if it rolls over, the crack was real and just arrived late. The checklist doesn’t hand you a verdict — it hands you the exact number and date that will.

Where the test misleads you — three false positives

The framework is sharp, but it has known failure modes. Watch for these before you trust a classification:

1. The “reset” that’s really a slow crack. A company can beat this quarter while the leading indicators quietly deteriorate — bookings slowing, backlog thinning, a key customer guiding down. The print says reset; the pipeline says crack. Fix: look one layer behind the headline beat at orders and remaining performance obligations, not just revenue.

2. The “hype run” that’s actually early truth. Sometimes the narrative is right and the number simply hasn’t printed yet — Nvidia in early 2023 looked like hype until the data-centre revenue arrived. Fix: a hype classification isn’t “short it,” it’s “don’t pay the narrative price until a number confirms it.” The error is buying the story, not disbelieving it.

3. The valuation reset that becomes a demand crack. A sentiment reset can turn into a demand crack if macro breaks underneath it — rates spike, a hyperscaler cuts capex — reclassifying the stock after you’ve already acted. Fix: the classification is a snapshot, not a permanent label. Re-run the five steps each print.

The discipline isn’t getting the box right once — it’s re-checking it every time a new number lands.

The week on one map

The five-step test sorts names into boxes; the map shows you why the boxes sit where they do. Plot every big mover on two axes — did the numbers improve (left to right), and how did the price react (bottom to top) — and the three boxes occupy different corners.

Read it once and the regime is obvious. Bottom-right — CrowdStrike and Broadcom — is the reset box: better numbers, lower price. Top-left — Marvell — is the hype box: a third of a trillion added on a quote, with no new number underneath. NVDA holds the top-right because its numbers are banked and undisputed. MU sits mid-bottom — it fell with no print of its own, the contagion case from the worked example. The empty corner is the one to fear: bottom-left, a real demand crack. Nothing’s there yet — and the four signals below are how you’ll know if something moves into it.

The Marvell tell: what pure sentiment looks like

Marvell deserves its own paragraph because it’s the purest specimen of the regime. The connectivity thesis behind it is genuinely sound: as AI compute disaggregates across thousands of chips, the value migrates to whatever links them — optical interconnect, switching, custom silicon. That’s a real, under-owned layer of the stack, and Marvell is a legitimate way to own it.

But the 32% one-day move — its largest ever — wasn’t triggered by a number. It was triggered by Jensen Huang calling Marvell “the next trillion-dollar company” on a Computex stage. Roughly $80bn of market value appeared on a sentence. Jim Cramer, no bear, called it “concerning… not based on anything other than one person saying it.” When a quote moves more value than most companies are worth, the price has detached from the print — and that detachment is the risk, regardless of how good the underlying thesis is.

This is the trap the new regime sets. A real thesis and a dangerous price can live in the same stock at the same time. The connectivity layer can be the right place to be and Marvell can be the wrong entry, simultaneously. The deep-dive discipline is to separate the layer you want from the price you’re asked to pay for it.

The harder question: reset, or the actual top?

Calling this a “sentiment reset” assumes the demand underneath is intact. That’s the bull case, and it’s a reasonable one — but it’s not free. The bear case uses the same facts: every late-cycle rally re-rates on sentiment before it rolls over, and “expectations caught up” is exactly what the top of a cycle feels like from the inside. Both cases are below.

My honest read: this is more likely a reset than a top, because the demand source is genuinely new and the prints are still strong. But notice what flips me: a single hyperscaler capex cut would move me to the bear column before any stock had “cracked” on its own numbers. That’s the one read-through I’m watching hardest, because it invalidates the bull case earliest. You don’t have to call it today — you have to know which signal would change your mind, and watch that one.

What to watch — four calendar-checkable signals

Each has a date and a binary read. None requires a forecast — only attention.

Turning a classification into a decision

A box on a map isn’t a trade. What it gives you is a different question for each name — and a threshold that has to clear before the question becomes an action. These are research postures, not recommendations.

A name in the reset box — the work is to separate valuation risk from fundamental risk. The business is intact, so the only question is whether the price is now reasonable or merely less unreasonable. A name on 39× forward sales that fell 9% is still on ~35×; “cheaper” isn’t “cheap.” The discipline is a confirmation threshold — a specific next-print number — and a size small enough to survive being early.

A name in the hype box — the work is to own the underlying layer without paying the narrative premium on one stock. If the connectivity thesis under Marvell is what attracts you, a broad basket holds that exposure while diluting the single-name sentiment risk. The threshold here is a printed number that converts the story into a fact — until then, the narrative price is the risk.

A name you can’t classify yet — the work is patience. The 24 June Micron call and the late-August CRWD/MRVL reports resolve the open cases for free. There is no cost to waiting for the print that moves a name out of the ambiguous middle — and a real cost to guessing before it does.

The rule that just changed

For two years the AI trade ran on one rule: own the names whose revenue points up, and let direction do the work. This week that rule broke. Direction is table stakes now — priced in, assumed, worth nothing on its own. What moves a stock from here is the gap between what it prints and what the market already believed.

The issue gave you the one-line version: did the business change, or just the price? This piece gives you what to do with the answer. “Just the price” isn’t one thing — it’s two: a reset (price fell, business intact) and a hype run (price rose, no number behind it) are both “just the price,” and they point in opposite directions. The three-box test is what separates them, and the reclassifying metric is what keeps you honest as each new print lands. A label you can’t update is just an opinion; a label with a date attached is a process.

Your next move — the classification exercise: take the three AI names you hold most of. Run each through the five-step test and place it in a box: reset, crack, or hype. Then do the part that separates investors from spectators — for each one, write down the single print, on a specific date, that would move it into a different box. A reset that loses its next ARR number becomes a crack. A hype name that finally prints becomes a reset. When you know what would reclassify a holding, you stop reacting to the price and start reacting to the business. That’s the whole discipline.

See you Friday.

— R. Lauritsen

Editor, iPrompt Signals

iPrompt Signals

Companion to iPrompt Signals Issue 13 · Published 5 June 2026 · FrontWave Media Ltd, Limassol, Cyprus

Disclaimer: for informational and educational purposes only. Not financial advice. iPrompt Signals is not a registered investment advisor. Always do your own research and consult a qualified professional.