This article was written by Miles Grauberd – Investment Analyst at I Know First.

The Early Cracks

On June 24, Micron posted the best quarter in its history. Revenue of $41.5 billion against a $35.9 billion consensus. Gross margin of 84.9%. Earnings of $25.11 per share. Management then guided the following quarter to $50 billion.

The stock popped 15%, ran to an all-time high of $1,255 on June 25, and has fallen 28% since. Micron broke below $900 and now trades below where it sat before the print. When numbers that good stop working, the business is not the issue. It’s the crowd. And the crowd started leaving memory before the tape admitted it. Our 1-year signal on MU was at 1,045 on June 4, three weeks ahead of the record print, with the stock near $1,000. Today that signal reads below 100. What followed was not a selloff. It was a handoff. Capital walked out of the AI hardware trade and into the software that runs on it, the security that protects it, and the banks that finance it. Below is where it went, why it stays there, and what the algorithm ranks now.

How The Trade Got Crowded

The thesis was simple. AI runs on GPUs, memory, and custom silicon. Software sells seats, and agents threaten seats. So the market bought the infrastructure and sold the applications. The trade worked for six months. That is how trades get crowded. By January 2, the spread between the semiconductor ETF and the software ETF hit the widest on record. Micron ran 250% year to date. Semiconductors grew into the single largest concentration in the index. Samsung and SK hynix went from a quarter of the KOSPI’s weight to more than half in six months. Hedge funds pressed until chips became the most crowded tech trade of the year. On the other side, Microsoft fell 23% in the first half, its worst start since 2000. June alone cost 17%. Twenty-two S&P 500 stocks more than doubled in the first half. The overlap among them was the trade. Concentration is not a risk until it becomes the only position in the room. By late June, hedge funds had begun unwinding the pair. They sold semis for four straight weeks.

Memory Broke First

The tell came before the print. On June 23, Korea’s Financial Supervisory Service warned on the leveraged single-stock ETFs tracking Samsung and SK hynix. Micron fell 13.3% to $1,051.77 on a regulatory comment about a product it does not sell, in a country where it is not listed. Nvidia dropped 4%, AMD 6%, Intel 6.1%. JPMorgan called it anxiety ahead of Micron’s earnings. Wedbush called it a gut check. Positioning was already the story. The earnings only confirmed it.

Micron beat by 16% on revenue, printed an 84.9% gross margin, and guided up. The stock rose 15%, then reversed the next session and closed down 6.69%. It has not recovered since. Our 1-year signal made a local peak at 1,045 on June 4 and now reads below 100, and the decay led both the high and the revisions.

The unwind spread, but not evenly, and that is the tell. The algorithm did not turn on semiconductors. It turned on memory. Nvidia carries the highest 1-year signal in the entire universe at 1,577 and has risen since the June low, while Micron sits at 88, Western Digital at negative 137. Since June 25, semiconductors have shed roughly $1.5 trillion in market value. Micron accounts for $350 billion of it. Sandisk, Intel, Applied Materials, and Lam Research each gave back more than $100 billion. Micron, Samsung, SK hynix, and the DRAM ETF all sit more than 20% below their highs, which puts memory in a bear market. Samsung posted a $59 billion operating quarter into that tape and the stock fell anyway. Korea carries the trade in leveraged form. On July 13 the KOSPI fell 8.95%, its seventh circuit breaker of the year, erasing 546 trillion won in a session. SK hynix fell 15.37%, its worst day on record. July 16th brought the 37th sidecar of 2026. Twenty-two S&P 500 names doubled in the first half. In the first week of July they fell 16.3% on average, and 20 of the 22 closed red.

Software and Cyber Caught The Bid

The capital did not leave the AI trade. It moved one layer up the stack. Over the eight sessions into early July, IGV gained 7% while SOXX fell 8.5%. In the week ending July 3, SOXX fell 4.84% and IGV rose 10%. The pair that defined the first half ran in reverse. Software rallied on valuation, not news. The sector spent the first half discounted on the thesis that agents replace seats, a story that hardened after February’s agent-tooling announcements. Microsoft fell 23% into it. Salesforce rose 4% and ServiceNow 6% on July 1 on upgrades alone.

Cyber ran harder. SOXX fell 6.6% over the six sessions off the June 25 low. CrowdStrike, Palo Alto, Zscaler, and Okta each gained more than 20% over the same stretch. CrowdStrike set an all-time high this week and Palo Alto sits just below its own, on a 25% one-month gain and 31% revenue growth.

The driver is budget. IBM said on Tuesday that clients were distracted with “rapidly-evolving, industry-wide cybersecurity concerns.” CrowdStrike +11%, Okta +10%, Palo Alto +7%. Cooling CPI did half that work, and CrowdStrike had no company news that day. But Gartner puts enterprise security budgets at $215 billion in 2026. Security survives a capex review, which matters once the market starts pricing return on AI spend rather than AI capex.

Financials Take The Baton

Technology slipped in the near-term sector table in early July. XLK fell 4.8% over the first five trading days of the second half, the worst sector on the board, and XLK/SPY printed its most extreme negative reading since April 2025. On the 14-day signal, technology now sits below financials and communications. Three-month momentum still favors tech at 30.4%. Rotations start at the margin, which is where this one is. Financials took the other side, and the quarter behind them explains why.

The banks reported through the week, and the pattern inside the results is the same rotation seen from the sell side. Trading revenue split cleanly in two. Equities roughly doubled while fixed income went nowhere. Equities revenue rose 86% at JPMorgan, 72% at Goldman, 70% at Bank of America, and 69% at Morgan Stanley, against 45% at Citi. Fixed income rose 6%, 9%, and 13% at three of them. The desks that got paid were the ones exposed to volatility in stocks, not rates, and the quarter delivered exactly that: the SpaceX listing, a re-rating in AI names, and a June that whipsawed the whole complex. Investment banking fees climbed 30% to 58% across the group, led by equity underwriting, where Goldman’s line rose 90% year to date on the back of secondary and initial offerings.

The driver is not AI. It is fee income and volatility, the two things a capital-markets franchise converts into earnings without needing a growth story. That is why the sector bid concentrates where it does. Citi is up 18.6% year to date and Goldman 15.4%, while Wells Fargo, with the least capital-markets exposure of the group, trades down 6.5%. The banks catching the rotation are the ones that get paid to intermediate it.

What’s Driving The Rotation

Four things, in order of how much they matter.

Positioning. Micron beat consensus by 16% and fell 28%. Samsung posted $59 billion in operating profit and sold off. On June 23, a Korean regulator’s comment about leveraged ETFs took Micron down 13.3%, in a stock that does not trade in Seoul. Hedge funds sold semis four consecutive weeks into early July, but the selling was surgical. Korea’s signal went negative while Taiwan’s held positive, memory versus compute, the same continent and the same AI trade splitting on what each side actually makes. When good news stops working and unrelated news does, the marginal holder is the story.

The AI-eats-software trade reversed. Software spent the first half discounted on the thesis that agents replace seats, a view that hardened after February’s agent-tooling announcements. Microsoft fell 23% into it. That de-rating built the setup. By late June the semis-to-software spread had reached a record and the discount became the catalyst.

The question changed from AI capex to return on AI spend. Meta guided 2026 capex to $125 billion to $145 billion, against $72.2 billion spent in 2025. Alphabet pays SpaceX roughly $920 million a month for compute. Numbers that size stop being a growth signal and start being a scrutiny magnet. Once the market prices what the GPUs earn rather than how many ship, the beneficiary moves from the supplier to the deployer.

Rates. The Fed holds at 3.50% to 3.75%. June CPI printed 3.5% against 3.8% expected with core flat, and June PPI fell 0.3%, its biggest monthly drop since April 2025, with May revised lower on both headline and core. Polymarket’s July hike contract collapsed to 4%, down 19 points on the week. But the same market prices close to a 60% chance rates sit higher by the October meeting, and Warsh told Congress the print was not mission accomplished. July came off the table. The year did not. Higher-for-longer compresses the richest multiples first, which is where the momentum names sat, and does the opposite to banks. That is half of why XLF is on this page.

The Signal Called The Rotation Before The Tape

In mid-June the algorithm ranked semiconductors well ahead of software and cyber. SOXX carried a 1-year signal above 600 on June 15 while IGV sat near 230 and HACK near 170. Then semis rolled over, and the signal rolled over first. SOXX peaked around June 15 and began falling while Micron still traded near its highs. The stock did not break until June 23, when a Korean regulator’s warning took it down 13.3%, and it kept falling through July 15 to below $900. The signal moved two weeks ahead of the price.

The order has now flipped. IGV leads at 270, HACK sits just behind at 183, and SOXX lags at 221 and falling. On the 14-day horizon the picture is sharper still: the algorithm reads SOXX at -0.24, outright negative, while IGV holds +0.29 and HACK +0.22. The near-term call is not that semis lag. It is that semis decline.

The individual names tell the same story. Micron, Lam Research, and Western Digital led the entire universe on the 1-year signal through spring. They now sit at the bottom: Micron 88, Lam 94, Western Digital negative at -137. CrowdStrike and Palo Alto lead, at 949 and 777, and CrowdStrike is the only ticker in the universe now reading above its early-June level.

The rotation into cyber has a fundamental driver behind it. On July 14, IBM’s CEO Arvind Krishna said clients were “distracted with rapidly-evolving, industry-wide cybersecurity concerns in the quarter.” Enterprises are treating cybersecurity as a first-order priority, and the spend that follows accrues to the names the algorithm already ranked at the top. I Know First saw it coming.

Risks To The View

The rotation is four weeks old. Semis still lead on three-month momentum at 30.4%, and XLK sits above its 50-day. Dip buyers have shown up for eighteen months. Microsoft and Apple report late July, and one hot print can unwind half of this in a session. The XLK/SPY reversal call that consensus made in early July failed twice, and tech deepened instead of bouncing both times.

The valuations on the other side are extreme. CrowdStrike trades at roughly 152x forward earnings and Palo Alto near 291x. Both just made or approached records. The bull case requires continued outsized growth surprises to hold those multiples, and the same rate sensitivity that lifted them on a cool CPI print cuts the other way. Cyber is not a defensive trade. It is a high-beta trade with a defensive story attached.

The memory story may be fundamental, not positional. This whole piece argues Micron fell on positioning. There is a competing explanation. IDC forecasts global smartphone volumes falling 13% in 2026, the largest drop on record, from a memory supply shock producers created by redirecting output to data centers. Micron broke $900 on reports that CXMT is expanding in DRAM. If pricing power is the real issue, memory is not oversold, it is repricing, and the software leg is a separate trade rather than the other side of the same one.

September is live. July hike odds collapsed to 4%, but the year did not. Roughly 60% of the market prices higher rates by October, and Warsh went out of his way to say the June print was not mission accomplished. Gas prices have already turned back up as the Middle East re-escalated. A hot August print compresses cyber and software faster than it compresses banks, which would leave one leg of this thesis standing and cost you the other two.

One number worth holding onto: the KOSPI is still up roughly 63% year to date after all of it.

Who Gets Paid

Micron will report again in late September. The quarter will almost certainly be enormous. That was never the question. Estimates went up and the multiple came down, which is what happens when the last buyer has already bought. The signal saw it on June 4, three weeks before the print that proved the business was fine and did nothing for the stock.

The money did not leave the AI trade. It moved to the layer that gets paid without carrying the memory cycle: the software running on the hardware, the security protecting it, and the banks financing the whole build. That trade is four weeks old and it can break. The rotation is not a call on AI. It is a call on who captures the value. The algorithm updates daily. So does the answer.