Memory Supercycle Escalates: Samsung's Record and Micron's $3,000 Target

Micron shares opened soft, sliding about 3% in the first five minutes on lingering unease over the Taiwan labor headlines, then staged a steady staircase recovery. The stock tagged roughly $1,090 intraday, closed near $1,088 for a gain above 4%, and probed the $1,100 resistance into the after-hours session. The presenter reads the tape as conviction buying: institutions and retail traders alike treating every dip in memory names as an entry.

Samsung prints a historic quarter

Samsung guided third-quarter operating profit to 107.4 trillion won, up 782.5% from a year earlier, becoming the first technology company ever to clear the 100 trillion won quarterly mark. As KoreaHerald reported, revenue also set a record at 195 trillion won, a result that puts Samsung in a club previously occupied only by an oil major and ahead of the $63.7 billion quarterly operating profit posted by the leading accelerator maker. Shares still slipped 1.77%, with macro nerves around Treasury yields outweighing the record print.

The preliminary release carries no divisional split, yet the analyst consensus is that memory did almost all the work. Estimates point to roughly 110 trillion won of operating profit from memory, a 75% margin for the semiconductor division and above 80% for memory alone. The phone and consumer-electronics side reportedly posted its first quarterly loss on record, so the headline number is really a story of memory earnings burying weakness elsewhere.

Two figures explain why. Reuters-tracked data cited in the video points to Samsung's high-bandwidth bit shipments rising nearly 50% quarter over quarter. And the TrendForce forecast relayed by Sedaily sees blended average selling prices climbing 121% next year as the mix shifts toward expensive new generations. Full divisional detail arrives October 29, which the presenter flags as the date that actually matters.

Why prices refuse to cool

The core thesis is that memory has left the classic cyclical frame and entered a structural supercycle. As conventional server memory gets expensive, it lifts the pricing floor for every high-bandwidth product; every wafer runs an internal bidding war against itself. Demand for low-power memory from autonomous software adds a second engine, leaving Micron's diversified portfolio positioned on the right side of the price action.

The second thread is tenser and comes from Taiwan. According to TaiwanNews, Micron's Taoyuan union won strike authorization with 99.1% support: more than 2,000 of 2,228 eligible members voted, turnout topped 89%, and 1,994 ballots backed action. The union demands a permanent, transparent profit-sharing scheme worth 15% of operating profit and expects a concrete plan from the board meeting on October 8-9.

Strike mandate in Taiwan

The dispute centers on the bonus gap. Under the 2025 plan, global payouts landed at 93.9% while Taiwan staff received 2.3 months of salary, later revised to 2.6 months. The chief executive's cash incentive equaled 32.4 months of base pay. On restricted stock , 11 million shares spread over 53,000 employees averages about 200 shares each, against 83,000 for the CEO, roughly 400 times the average; the union also cites 225,000 shares granted in loss-making 2023.

No strike date has been declared and the fabs keep running, so the vote is best read as a high-stakes bargaining chip. The real danger, the presenter warns, is a fab walkout: it would send spot prices through the roof and hand Samsung and SK rivals pricing leverage over the largest cloud buyers. That overhang explains the early selloff, and the October 8-9 talks are the near-term clearing event. The presenter adds his own contested view that the CEO's direction-setting justifies higher pay.

The loudest headline is D.A. Davidson analyst Gil Luria lifting Micron's target from $2,100 to $3,000. CNBC reports the new target implies 187% upside from the $1,045.56 close and rests on a simple claim: investors are still early in understanding the company's worth, with memory demand set to outrun supply for years. The note's punchline, that if retail will not buy it institutions will, captures the street's mood.

The math behind $3,000

The arithmetic is a multiple bet, not an earnings bet. Luria applies 19 times estimated fiscal-2027 earnings per share of $158; the stock trades near 6 times that year's earnings today, with a trailing multiple of 14.4. Consensus earnings of $172.77 sit close to Luria, so the fight is over the multiple. Bulls cite constrained supply and rising estimates; bears answer that margins near 87% leave no room, consensus flattens at $49-51 a quarter through 2028, and 19 times would break the historical pattern after a 1,500% run.

The funding side of the cycle got its own blockbuster. Investing reports a space and AI company seeking about $40 billion, split into $10 billion of bank loans and $30 billion of investment-grade debt, to buy accelerators, with closing expected in 2027. Apollo is set to lead the financing and Pimco sits among the lenders; the buyer has committed to building exclusively on the newest architecture. Every accelerator installed pulls memory and enterprise storage with it.

Custom silicon confirmed the breadth of the buildout at Marvell's investor day. StreetInsider reports fiscal-2028 revenue guided to about $20 billion, up from a $15-18 billion band, with a $70-90 billion range for fiscal 2031 against a $400 billion addressable market by 2030. Multi-year programs with large cloud customers plus a $2 billion investment from a major accelerator maker underpin the frame. A second calendar catalyst sits in December, when post-grant restrictions fall away and a large buyback around December 9 becomes possible.

Stepping back, the signal and the noise point opposite ways: the noise is Taiwan, the signal is deepening scarcity. Samsung's October 29 detail, Micron's October 8-9 board talks and the December buyback window form the three markers that matter into year-end. Whether $1,100 turns from ceiling to floor depends on memory earnings holding up and the labor file closing quietly. The open question is no longer where the cycle stands, but how much memory weight portfolios should carry.