The Chip Selloff, Measured

The Setup

Last Friday, in Where the Rotation Went, we followed the money out of semiconductors and into the banks that caught it. This week we walk back through what the money left behind, because the damage kept compounding: the chip index is now down 17.9% since late June, Micron has shed more than a quarter of its value from its high, Qualcomm roughly 40% from its peak, and the selloff's stated cause, doubt about the durability of AI capital spending in a hawkish-Fed world, remains unresolved after Wednesday's divided hold. Then came Thursday: Samsung reported a historic quarter and said the memory shortage could persist through 2028, and the complex bounced hard, led by an 18% single-day recovery in Micron. A month of wreckage and one violent rebound is exactly the setup that turns everyone into a dip buyer. It is also exactly when measurement matters most.

So we measured. Every tracked name in the semiconductor complex, 39 in all, spanning accelerators, memory, analog, handset silicon, equipment, and foundry, pulled through the platform's three scoring systems as of this week. Standing exclusions applied: Taiwan-listed names whose fair values print with currency artifacts came out as directional-only, and any name whose calculated upside exceeds our 80 to 100 percent reliability band came out on data integrity. Then one organizing fact emerged from what remained, and it built this list: through the entire month-long drawdown, the platform's outlook engine stayed Bullish on exactly six of the 39. Everything else reads Neutral or Bearish. The market sold the whole sector; the scores kept conviction in a handful.

The five below are those Bullish survivors ranked by valuation, from the deepest discount to the one name that carries conviction without cheapness. The sixth survivor missed on price alone and leads the next-in-line section. What this list is not: a claim that the bottom is in. Outlook streaks are built from data that includes momentum, and momentum can break. What it is: the platform's honest answer to the week's question. After a 17.9% sector decline, how much genuine value did the wreck actually create? Less than the bounce suggests, more than zero, and concentrated in names the AI narrative mostly ignores.

Cirrus Logic (CRUS)

Cirrus Logic (CRUS) tops the list, and regular readers will recognize the name: we published the full Investment Thesis on June 30, four sessions before the sector broke. The audio and mixed-signal specialist, whose amplifiers, codecs, and power converters sit inside essentially every flagship smartphone, has since done what quality does in a storm: fallen 9.9% against the sector's 17.9%, holding its relative ground while the complex came apart around it. The scores held too. Company Strength of 82 remains Elite, the second-highest rating in the entire 39-name scan, on Financial Health of 88/100, the best in the scan outright, with a Moat of 13/15 and a debt-light balance sheet the June article covered in detail.

What the selloff changed is the price of all that. Fair value calculates to $158.25 against a price near $132, a 20% discount that makes Cirrus the single largest clean value in the semiconductor complex right now, with a Bullish outlook in eight straight readings. The June thesis argued the market was underpaying for a company whose engineering position inside its customers' devices functions as a moat regardless of which phone model wins; a month and a sector crash later, the platform's math says the underpayment got larger while the business case did not change.

The known weakness has not changed either, and we restate it every time: customer concentration. Cirrus earns the overwhelming majority of its revenue from Apple, which makes it structurally a leveraged bet on one relationship, and no strength score can diversify that away. The June article treated this as the central risk, and the intervening month cut both ways on it: the memory-cost squeeze now hitting smartphone economics, which we will meet again two sections down, touches Cirrus's end market too. Elite quality at a 20% discount is the scan's best offer. It comes, as it did in June, with a single point of failure attached.

Micron (MU)

Micron (MU) is the epicenter of everything this article is about: the supercycle that carried the sector up, the doubt that dragged it down, and the Samsung report that bounced it Thursday. The memory maker crossed $1 trillion of market value in May, has risen more than 700% over the past year, and printed fiscal second-quarter revenue of $23.9 billion, up 196%, with guidance for $33.5 billion next quarter at 81% gross margins, numbers with no precedent in the history of the memory industry. The engine is high-bandwidth memory for AI systems, effectively sold out through 2026 under binding multi-year contracts. Then July repriced the doubt: the stock fell 27.9% from its $1,213 high before Samsung's Thursday report, signaling record memory economics and a shortage that could run through 2028, sent it up roughly 18% in a session.

One disclosure before the scores, because our rules require it and the story does too: Micron is a commercial partner of, and investor in, Anthropic, the maker of the Claude AI that powers portions of the Wealth Engine Pro platform, supplying memory and storage for Anthropic's systems and co-developing future components. We flag that relationship inline whenever it exists, and the full statement sits in the disclaimer below. Now the platform's read: Company Strength of 67, Strong, on Financial Health of 72/100 and a Growth score of 13/15, with fair value at $1,016.53, a 16% discount to the price near $875, and a Bullish outlook in eight straight readings that never wavered through the drawdown.

And now the caveat, which for this name deserves its own paragraph, because memory is the most cyclical business in all of semiconductors and the platform's models cannot tell a supercycle from a cycle. That $1,016 fair value is computed from earnings that include 81% guided gross margins, margins that exist because three manufacturers cannot meet AI demand; the contracted backlog argues this cycle is structurally different, and the entire history of the memory industry argues that nothing is ever structurally different. Two honest yellow flags belong on the record alongside the Bullish streak: Micron's chief executive sold roughly $37 million of stock on Monday, and a June antitrust suit over DRAM pricing adds a legal overhang the models do not price. The discount is real. It is a discount to supercycle earnings, and the reader deserves both halves of that sentence.

Qualcomm (QCOM)

Qualcomm (QCOM) is the crash on this list, and the irony at its center is worth the price of the whole scan: the handset chip giant is being squeezed by the very shortage making Micron rich. The stock has fallen roughly 40% from its 52-week high of $259.92, hammered by a June strategy pivot toward AI data center silicon that drew a wave of downgrades, by Apple's ongoing transition to in-house modems, and above all by what management calls unprecedented increases in memory prices, which are crushing the economics of the phones Qualcomm's chips go into. Wednesday's report made the squeeze explicit: revenue of $9.95 billion beat estimates, but fourth-quarter profit guidance came in about 9% below consensus, and the company said Android handset revenue will fall roughly 20% this fiscal year, costing more than $1.50 in earnings per share. Qualcomm's response starts September 1: double-digit price increases of its own.

Here is what makes it a list entry rather than a cautionary tale: the platform's models say the 40% decline took the price almost exactly to fair value, $156.97 calculated against roughly $152 on the tape, and the outlook engine read Bullish through every leg down, eight for eight. Company Strength holds at 66, Strong, on Financial Health of 78/100, the second-best on this list. The stock now trades near 14 times forward earnings with a 2.4% dividend yield, and the average analyst target sits about a third above the price even after six downgrades, the rare configuration where the sell side capitulated faster than its own spreadsheets.

The forward question is whether the squeeze is cyclical or structural. The bear case stacks real things: memory costs, the Apple attrition that has loomed for years, and a capital-intensive pivot into data center markets where incumbents are entrenched. The bull case is that handsets bottomed, China handset revenue is guided back to sequential growth, automotive and IoT now supply 40% of chip revenue, and the just-closed $3.9 billion Modular acquisition brings one of the industry's most respected AI software teams to the pivot. The platform takes no side on the story. It says the price, for the first time in this cycle, no longer requires one.

NXP Semiconductors (NXPI)

NXP Semiconductors (NXPI) is the quiet one, which is itself the finding. While the AI names crashed and bounced, the Dutch automotive and industrial chipmaker reported a Tuesday quarter that beat on every line and barely made a headline: revenue of $3.5 billion, up 19% and growing in all four segments, adjusted earnings up 33%, operating margin expanding three points to 35.1%, and free cash flow of $791 million, nearly 23 cents of every revenue dollar. Automotive, the core at $1.9 billion, grew on rising chip content per vehicle rather than inventory restocking, which management explicitly said it is not seeing. Industrial and IoT rose 38%, communications infrastructure 41%, and guidance calls for 18% growth again next quarter with data center revenue more than doubling this year to over $500 million.

The platform prices all of that at almost exactly what it costs: fair value of $241.64 against a price near $245, within about a percent, with Company Strength of 61, Moderate, and a Bullish outlook in seven of the last eight readings, the only name on this list without a perfect streak. That combination, accelerating fundamentals at a fair price with improving conviction, is the profile of a business the market has neither punished nor discovered. NXP's end markets, cars, factories, and infrastructure, sit far enough from the AI capex debate that the selloff mostly passed it by, and far enough from the AI narrative that the mania did too.

What holds it to fair value rather than a discount is honest: autos are cyclical, the growth-driver framing leans on a software-defined-vehicle transition that will take years to fully monetize, and a company guiding to second-half strength is making a forecast the macro can still veto. But on a list bracketed by a supercycle and a crash, there is a role for the name whose entire story is execution at a reasonable price. The scores say that is exactly what this is.

Nvidia (NVDA)

Nvidia (NVDA) closes the list because leaving it out would dodge the question every reader of a chip-selloff article is actually asking. So here is the platform's full answer on the AI leader, unrounded. Quality: the highest Company Strength in the entire 39-name scan at 85, Elite, with Financial Health of 84/100 and Moat and Growth both at 13/15, numbers that describe exactly what Nvidia is, the most dominant franchise in the sector and one of the most dominant in the market. Conviction: a Bullish outlook in eight straight readings that did not blink through the drawdown. Price: 25% above the platform's calculated fair value of $146.21, even after falling 13.1% from its June high.

Read that as the two-handed verdict it is. The selloff brought Nvidia closer to its fair value than the premium it carried at the June highs, and the scores never lost conviction in the business. But closer is not there: a reader who buys at $195 is paying a quarter more than reported financials, formidable as they are, can currently justify, which means paying for continuation, for the capex cycle whose durability is the precise thing this month's selloff called into question. The equipment makers two sections down show what the far end of that bargain looks like; Nvidia is the near end, the smallest quality-adjusted premium the AI trade's core has offered in a long while, and still a premium.

Its inclusion also keeps this list honest about what the scan found. Ranked purely by valuation among the six Bullish survivors, Nvidia is fifth, ahead only of a name that missed the cut entirely. The measured conclusion is not that the king is cheap. It is that the king is the best business in the sector, priced 25% past its filings, falling toward the number rather than away from it. When those two lines cross, the platform will say so on its own. They have not crossed.

What the Wealth Engine Scores Say

Here is the platform's systematic read on all five, current as of Thursday. These are five of the only six Bullish outlooks the 39-name scan produced, ordered from deepest discount to widest premium.

Notice the shape of the list. The valuation column runs in a straight line from a 20% discount to a 25% premium, while the outlook column never changes, and the discounts sit on the names the AI narrative treats as supporting cast: an audio specialist, a handset chipmaker in a squeeze, an automotive supplier. The premium sits on the star. A month-long, 17.9% sector decline redistributed prices without redistributing the story, and the scores, which cannot hear the story, priced the difference.

These scores are systematic. They evaluate companies on reported financials, balance sheet quality, moat characteristics, and valuation models including discounted cash flow, peer comparison, and earnings power. They measure what a company is today, not what it might become. In this sector right now, that cuts differently name by name: Micron's fair value inherits supercycle margins and carries our written cyclical warning, Qualcomm's inherits a squeezed year that may mark a bottom, and Nvidia's inherits financials that, extraordinary as they are, still trail what the price assumes comes next.

The outlook streaks deserve their own sentence, because they are the organizing fact of this edition: through a drawdown that cut the sector by nearly a fifth, 33 of 39 names lost the engine's conviction and these five, plus one, did not. That is not a prediction that the selloff is over. It is a measurement of where the data's conviction and the market's panic diverged, and divergences are where this publication does its work. Pull up any of the 39 on the platform and check the gap yourself.

What Didn't Make the List

The sixth Bullish survivor was Analog Devices, and its exclusion is the valuation rule working exactly as designed: Strong at 75, an eight-for-eight Bullish streak, and a price still 40% above calculated fair value even after the selloff. Conviction without cheapness ranked it sixth of six, and a five-name format cuts at five. It is the first name to research if this list feels one short.

Super Micro is the opposite lesson and worth pausing on. Three weeks ago its calculated upside was 88%, just past our reliability band; the stock kept falling and the figure has ballooned to 166%. That is not opportunity growing, it is a model breaking in public, and it is precisely why the band rule exists: past a certain distance between price and model, the number stops measuring the stock and starts measuring the model's blind spot. The same standard excluded Taiwan Semiconductor and UMC, whose fair values print with currency artifacts, which pains us in TSMC's case: it carries an Elite 80 strength score and an eight-for-eight Bullish streak, and deserves a clean valuation read our models cannot currently give it.

Then the tiers the scan flagged, named without ceremony. The equipment oligopoly is the selloff's starkest non-event: ASML, Lam Research, Applied Materials, and MPWR all carry Elite strength scores, and all still trade 45% to 71% above fair value after the drawdown, with KLA at 74%; the toolmakers' premiums barely dented. At the bottom, the engine turned Bearish on Microchip, Silicon Labs, Allegro, Camtek, and Onto, names falling and flagged simultaneously. And Intel, after last week's report, covered in our Avoid Thesis, still reads Bearish eight-for-eight at 64% above fair value, a reminder that a crashing price and a cheap stock are different things.

What Could Go Wrong

Start with the instrument this list is built on. An outlook streak is not a floor; the engine's conviction incorporates momentum and revisions, and both can break after publication, exactly as they broke for the 33 names that lost their ratings on the way down. Buying Bullish survivors of a drawdown is a bet that the drawdown was a correction inside a cycle rather than the top of one, and Wednesday's Fed hold resolved nothing about the question that started the selloff: whether AI capital spending, the demand signal underneath every name here except NXP, holds its trajectory into a higher-for-longer world. Thursday's Samsung bounce was one data point. Bear markets are full of 18% up days.

The two value names carry mirror-image hazards. Micron's discount is computed against supercycle earnings, and memory has broken every structurally-different argument ever made about it; if Samsung and SK Hynix answer 81% margins with capacity, the fair value deflates with the margins, and the chief executive's $37 million Monday sale plus a pending DRAM antitrust suit are the kind of details that matter more in hindsight. Qualcomm's fair value is a photograph of a squeezed year, which protects it from extrapolating a boom but not from the squeeze deepening: memory costs are rising into its September price increases, not after them, and the Apple modem attrition is a scheduled subtraction no recovery elsewhere cancels. Cirrus concentrates that same smartphone exposure into a single customer relationship, Elite scores notwithstanding. And NXP's fair-value tag assumes an automotive second half that a hawkish Fed, a squeezed consumer, or a stalled software-defined-vehicle transition could each individually spoil.

Finally, the honest frame around the whole exercise: this is a list drawn from the wreckage of the market's most crowded trade, published the day after its biggest bounce in months. If the rebound runs, the discounts above will close and this edition will look prescient; if the selloff resumes, Bullish streaks will snap and we will report which ones, by name, because that is what the streak data is for. The scores measured where conviction survived the first leg down. Nothing in them promises there is no second leg.

The Bottom Line

Measure the chip selloff and the shape of it surprises you. A month of indiscriminate selling turned out to be discriminate after all: the outlook engine kept its conviction on six names out of 39, and the value it created pooled almost entirely in the sector's unglamorous corners. The single best offer in the complex is an Elite-rated audio specialist at a 20% discount that we happened to publish the case for four days before the break. The supercycle at the center of the storm trades at a discount to earnings the models cannot promise will last, and says so in writing. The crash landed a squeezed handset giant precisely on fair value. The quiet automotive compounder never left it. And the king, the name the whole debate orbits, fell 13% and remains a quarter above what its filings support.

That distribution is the answer to the week's question. Did the selloff put the sector on sale? Mostly, no: the equipment oligopoly kept premiums of 45% to 74%, the AI leader kept a quarter, and the names that got genuinely cheap are the ones the AI story was never about. Anyone buying the dip in the sector is, in the platform's math, mostly buying smaller premiums, not discounts. The exceptions are five names long, and each carries its warning label in the section above.

That is the Wealth Engine Pro approach: when the market has its most emotional week, run the largest scan, publish the streaks and the artifacts alike, and rank by the numbers rather than the narrative. Next Friday the Sector Spotlight rotation continues with healthcare, and these five join every list before them in the public record, where the scores will grade what the bounce and the Fed and the memory cycle do next. The dip buyers have a story. As of Thursday's close, this is what it measures.