If More Than Half of 2027–28 Memory Is Already Sold, Where Does Cash Stick?

ANALYSIS · MEMBERS CONTINUE · 5 min free / 14 min full · Data as at 28 Oct 2026

On 28 October 2026, SK hynix printed the first ₩100 trillion sales quarter in its history — ₩104.6155 trillion of revenue, ₩83.3338 trillion of operating profit, an 80% margin on the company's rounding — and then told the same room two things that pull in opposite directions. It will hand about ₩100 trillion back to shareholders this year. And long-term agreements with 14 customers already cover more than half of the memory it can make in 2027 and 2028. The new lines that would loosen that book are still a ramp. This map asks who is paid if more than half of 2027–28 memory is already sold.

In short

• The question: if more than half of 2027–28 memory output is already sold, where does the cash stick?
• The answer: Scenario B. The sold book, not the cheque, is the stuck step. The 14 LTA customers get the allocated stacks. SK hynix is paid an 80% margin on what shipped. Shareholders are paid ₩100 trillion — 1.01× net cash and 1.20× this quarter's operating profit (ours). Everyone else competes for the remaining under-half of two years. That remainder is not a queue
• Why: this morning's Wire gave the cheque and the >50% book. Today's Korea Pulse kept HBM4 as the bind. Today's Stack shows Tokyo Electron already paid on coater-tracks (Korea 20.8% of April–June) while 2027 capex stays unnamed. This map is the sold book, not a stock
• The size of the gate: remaining open capacity is less than half of 2027 and of 2028. HBM customer ask is more than twice what the company can ship (call). Those are two gates, not one ratio. We do not invent a wafer-start. M15X is still a year-end "maximum," not a cube print. Yongin Y1's first cleanroom is dated February 2027
• What would break it: a named M15X or Yongin wafer-start that actually fills the under-half (A), or a later call that prints the "sold out through 2029" line as signed LTAs rather than LTAs plus demand still being discussed (C)
• Nothing here is a recommendation. Arithmetic shown below the wall

The numbers that matter

Sources: SK hynix newsroom and earnings call, 28 Oct 2026; SK hynix, 7 Aug 2026; Tokyo Electron tanshin, 30 Jul 2026. Ratios are ours. Consensus is press, not a company print.

Why is the sold book the stuck step?

An AI accelerator does not leave without high-bandwidth memory beside the logic die. HBM, explained is the free page for the stack. Korea still invoices a large share of those stacks. This morning SK hynix told the room that the stacks it can make in 2027 and 2028 are already more than half spoken for — before M15X is at maximum, before Yongin Y1's first cleanroom, and before the company has numbered next year's investment.

That is a different question from 1 October, when HBM taking ~30% of DRAM wafers mapped wafer mix, not a two-year LTA wall. It is not what SK hynix's price assumes about the HBM4 back end. It is the inverse of cash booking the finish before the line is live: there cash had left and the tester waited; here the seller has already sold the wafers. Micron locked more than 75% of fiscal 2027 — one year. Today's print is more than half of two years. The call's "sold out through 2029" line stacks LTAs with demand still being discussed. We map the printed >50%.

Three clocks around it are not this bind. The P&L already cleared: ₩83.3338 trillion of operating profit on the stacks that shipped in July to September. The cheque to the register is a decision about cash that followed that margin, not a cube. The coater-track at Tokyo Electron is already an invoice — ¥531.3 billion of SPE new equipment in April–June, Korea 20.8% of the quarter — and it still sits beside an ASML slot. TEL's next print is 30 October against a ¥1,620 billion first-half guide. That is a tool clock. None of those three turns an unsigned 2027 wafer into a free one. What is left is a sold book: 14 names, more than half of two years, deposits so plans become orders, prices that still move inside a band. The under-half is what every buyer without an LTA will compete for. HBM ask more than twice supply is the second gate on the same door. We do not divide 2.0 by 0.5 and call it a company figure.

What does the map already assume?

Three answers compete: M15X and Yongin Y1 convert the under-half into open supply inside a year (A); cash sticks with the 14 LTA customers and with the register, while the remaining under-half of 2027–28 stays the bind (B); or more names sign, terms extend, and the "sold out through 2029" line becomes the printed LTA book rather than LTAs plus discussion (C). A tip sheet would pick the stock. Analysis asks which answer this morning's >50% already needs.

Our read, in one line: the map already assumes Scenario B. More than half of 2027 and of 2028 is already sold. ₩100 trillion is 1.01× the ₩98.7 trillion net-cash pile and 1.20× this quarter's operating profit (ours). HBM ask is still more than twice supply. M15X is a year-end maximum, not a named extra wafer. Tokyo Electron is paid on tracks that already shipped; it is not paid on an SK hynix 2027 capex figure that this morning still did not print. This is not a company price. The scenarios and the arithmetic are below.

The free preview ends here.

—— Still to read · ~9 of 14 minutes ——

MEMBERS CONTINUE HERE — who gets paid · one sold year · three scenarios · takeaways · FAQ