Between Friday's close and Wednesday morning, Micron traveled from $848.95 to $974, a 14.8% repricing of a trillion-dollar company in three trading sessions. Here is the complete list of changes to the business over those three sessions: a UBS note arguing the company will generate more than $400 billion of free cash flow through 2028, and a record Taiwan export order print confirming the demand pipeline. Both positive. Everything else was mood.
I own a large Micron position, and I own it the same way at $848 as at $974, because I own the cash flows, not the chart. But a week like this one deserves an honest writeup, both for what it says about the asset and for what it taught about holding through a panic that briefly tripped my own published risk line. Fundamentals first, valuation second, trend last, in exactly that order. Here is the week through that lens.
The week, precisely
Wednesday, July 16: TSMC printed the best quarter in its history. Revenue $40.2 billion, up 34% year over year at the exact top of guidance, gross margin 67.7%, net profit $22.4 billion against $19.7 billion expected. It raised 2026 capex to $60 to 64 billion from $52 to 56 billion. The market read the capex raise as an industry-wide free cash flow warning and sold every supplier: SOX down 2.6%, $MU down 5.65% to $853.20, $SKHY and $SNDK down 8%.
Thursday night: Moonshot AI released Kimi K3, a 2.8 trillion parameter Chinese model with a 1 million token context window, and the reflex trade called it DeepSeek 2.0. Nasdaq futures fell 2%, the Asian semi complex dropped 6%, and Apple passed Nvidia as the world's most valuable company. Micron finished that week at $848.95, briefly below the $860 risk line I published July 7.
Monday, July 20: UBS published the note of the month, more than $400 billion in free cash flow through calendar 2028, with capacity to repurchase over 40% of shares outstanding once the buyback restriction expires December 9. The stock closed back above the line at $865.46.
Tuesday, July 21: up 12.2% to $970.82 as the narrative flipped from "memory is over" to "AI memory boom" inside three sessions. Taiwan reported June export orders up 59.4% year over year to a record $95.26 billion the same morning, electronics up 79.9%.
Wednesday, July 22: $974 at midday, $1.1 trillion market cap, Alphabet and Tesla reporting tonight.
Two panics that inverted on contact with the fundamentals
The TSMC panic priced capex as a cost. For the companies supplying the buildout it is revenue. Every AI accelerator TSMC packages ships with HBM stacked beside it, and TSMC raised its own revenue growth guide to 40% from 30% in the same breath. Selling memory because the foundry at the center of AI is building more capacity was first-order thinking, and it survived less than three sessions.
The K3 panic was built on an efficiency-shock template the launch sheet contradicted. K3 launched at $3 per million input tokens and $15 per million output, roughly triple its predecessor's pricing. Efficiency shocks do not arrive with price increases. A 1 million token context window is, mechanically, a memory bill: long-context inference lives in DRAM and HBM. The model that triggered a memory selloff is one of the most memory-hungry products ever shipped. There was a credibility tell too: K3 kept introducing itself as Claude in user screenshots, which is what training heavily on a rival's outputs looks like. Following the frontier is cheap. Funding it is not, and that funding is exactly what the selloff was doubting.
Neither panic touched a single Micron number. That is the part a long-term owner actually needs to verify, so I did.
What I own, and what it did this week
The investment case is cash flows under contract, and the week strengthened every leg of it. HBM remains sold out into 2027, with 2027 HBM4 contract pricing still tracking toward roughly double current levels per Korean industry press. The take-or-pay contract base, roughly $100 billion through 2030 backed by $22 billion of customer deposits, puts price floors under about 40% of revenue; Micron added supply agreements with seven automotive customers, including Qualcomm, DENSO, and Hyundai Mobis, on the worst day of the selloff. Fiscal 2027 consensus EPS sat at $149.64 before the whipsaw and sits there after it, untouched by two separate panic days. Taiwan's export orders, the physical order book for the whole AI supply chain, printed an all-time high for June and an all-time high first half, $504.1 billion, up 50.9%.
Then the ownership math. If UBS is even half right about $400 billion of free cash flow through 2028, a company valued at $1.1 trillion could retire a fifth to two fifths of itself starting December 9, when management has already committed to returning 100% of excess free cash flow. At $974 the stock trades near 6.5x fiscal 2027 consensus earnings. I can find things to argue with in that setup. Deterioration is not one of them, and nothing that happened between Wednesday and Wednesday changed a line of it.
What the risk line is actually for
I publish one trend rule on this position, a weekly close below $860, and Friday's $848.95 close technically tripped it before Monday reclaimed it. So let me say plainly what that rule is and is not. It is not a trading system, and it did not make me a dollar this week. It exists because I am a concentrated owner of a 2-beta stock, and a concentrated owner needs one pre-committed answer to the question "at what point does the market know something I don't?" written down before any panic, sized so that even a violent break would be survivable rather than forcing.
The week showed both edges of it. Mechanically dumping the position on Friday's close would have sold the low of the month, 13% below Wednesday's price, on a close that came mid-reversal with the business data actively improving. Having no line at all would be worse discipline, not better; some week the break will be real. The rule's actual job was done quietly: it kept one ugly session from forcing a decision the fundamentals never endorsed, and it kept me honest in public while the tape whipsawed. The fundamentals gate decides if I own Micron. The line only decides how carefully I watch it.
The next five sessions
The calendar does the underwriting from here. Tonight, Alphabet and Tesla report, the first mega-cap capex commentary since the K3 panic. Thursday, Samsung's full second quarter and Intel, the real memory read of the week. Friday, Micron's July options expiration and the weekly close. Monday, CXMT begins trading in Shanghai, the first market price on the China memory threat I broke down in the last piece. Four fundamental data points in five sessions, each one testable against numbers already published here.
My position is unchanged: long, not trimmed through either panic, underwritten by contracted cash flows at 6.5x forward consensus, with one written-down risk line I will keep honoring in public whichever way it resolves.
Disclosure: I am long MU. This is my own research and opinion, not financial advice. Do your own work, especially with a 2-beta trillion-dollar memory stock reporting catalysts four days in a row.