Goldman keeps its KOSPI target at 12,000 (September 7) — the index trades at 5.3 times forward earnings

The three lines

• Target — KOSPI 12,000 maintained. Timothy Moe, Goldman's chief Asia-Pacific equity strategist, on Bloomberg TV, September 4
• Basis — 7.5 times twelve-month forward earnings against a current 5.3, described as half the seven-year average
• Number — US big tech capex seen revised from 800 billion to 1.2 trillion dollars. From September 7's close of 6,995.39, the target implies about +72 percent

Key questions

What is the basis for Goldman's 12,000 KOSPI target

**A single valuation multiple — 7.5 times twelve-month forward earnings.** Timothy Moe, Goldman Sachs's chief Asia-Pacific equity strategist, said on **Bloomberg TV on September 4, 2026** that the bank is holding the target; Korean outlets carried it on **September 7**. The arithmetic is plain: **① estimated earnings for the next twelve months** multiplied by **② the multiple the market pays**. Goldman uses **7.5** for ②. **The index currently trades near 5.3**, which Moe called about **half its seven-year average**. **What matters here is that the forecast is not built on raising earnings estimates.** Most of the implied gain comes from assuming the market will **pay more for the same estimates**. With the KOSPI closing at **6,995.39 on September 7** (「KOSPI closes at 6,995.39 on September 7」), 12,000 sits about **72 percent** higher. For scale, the index's record close is 7,384.56, set in May 2026 — meaning the target is well above any level Korean equities have reached.

Why is this not a peak-out

**Because demand estimates are rising while supply cannot follow.** Moe's central number is **US big tech capital expenditure for next year**, which he expects to be revised from about **800 billion dollars** to **1.2 trillion** — an increase of **400 billion, or 50 percent** (「What AI capex is」). His second point is that **the memory shortage created by data-center competition will deepen next year**, because production capacity is not keeping up. **Understanding why 'peak-out' is the contested word requires the industry's history.** Memory has long been read as a business where every boom ends in oversupply: prices rise, everyone expands, and prices collapse when the new capacity comes online. Moe's diagnosis is that the market is pricing that cycle again, and his rebuttal is that **this one runs longer** (「What a memory supercycle is」). The physical basis for the argument is lag — the time between a capital-spending decision and actual output reaching the market (「How long semiconductor capex takes to reach prices」). Korea's own trade data pointed the same way the same week: cumulative 2026 exports reached 709.4 billion dollars by September 5, passing all of last year, driven overwhelmingly by semiconductors (「Korea's exports pass 709.4 billion dollars (September 5)」).

How much weight should this carry

**Read it as two conditions that both have to hold.** **Condition ①: earnings estimates are met.** Moe attached this himself — the target does not look unreasonable **if Korean companies deliver their own earnings forecasts.** **Making delivery a condition means that a shortfall lowers the target regardless of what the multiple does.** **Condition ②: the multiple rises from 5.3 to 7.5.** A multiple is not an earnings number; it is **the market's disposition.** What investors will pay for the same profit is set by rates, currency and geopolitics — things outside the companies. The KOSPI's recent behaviour shows how exposed that is. On September 2, Samsung Electronics and SK hynix fell about 4 percent and the index dropped 273 points (「KOSPI closes at 6,562.72 on September 2」). On September 7 the same two rose 5 to 8 percent and pulled the index up 4.61 percent. **A rally built on one axis unwinds on the same axis.** Finally, **this is a target, not a dated promise.** Goldman did not say when 12,000 would be reached, and the Korean reporting carries no horizon.