Taiwan-Korea stock valuation gap hits record high as market bets on durable semiconductor bottlenecks

Taiwan gains a semiconductor premium as AI demand shifts from memory cycles toward durable bottlenecks in logic, packaging and HBM markets.

A price-to-earnings (P/E) multiple is a forecast written in shorthand. It says less about what a company earns today than about how long investors believe those earnings will survive competition. Read that way, the most revealing number in Asian equities this autumn is a ratio.

Taiwan's benchmark TAIEX (Taiwan Capitalization Weighted Stock Price Index) trades at about 18 times its one-year forward earnings. South Korea's KOSPI (Korea Composite Stock Price Index) trades at about 5.5 times, according to market estimates reported on October 7, 2026. Both markets ride the same artificial intelligence (AI) hardware boom. The gap between them is a market judgment about which form of semiconductor scarcity will prove more durable.

That judgment is striking because Korea's numbers have rarely looked stronger. On October 1, Korea's Ministry of Trade, Industry and Resources (MOTIR) reported September exports of US$120.94 billion, up 83.5 percent from a year earlier and the first month above US$120 billion. Semiconductors alone reached US$60.30 billion, up 262.8 percent, and memory chips accounted for US$54.07 billion of that. Samsung Electronics subsequently estimated third-quarter operating profit at ₩107.4 trillion (about US$80.2 billion at the company's reported exchange-rate conversion), nearly nine times the ₩12.17 trillion reported a year earlier and slightly above the London Stock Exchange Group (LSEG) SmartEstimate of ₩106.1 trillion based on 21 analysts.

Yet KOSPI fell 19.3 percent in the July–September quarter while TAIEX rose 3.9 percent. That 23.3-point gap is reportedly the widest since the turn of the century. By the close on October 7, TAIEX stood at 49,806.37, up 72.0 percent for the year, according to Taiwan Stock Exchange data. KOSPI stood at 6,803.90, up 61.5 percent, but 25.4 percent below its June 22 peak of 9,114.55.

Chart 1 Caption: TAIEX and KOSPI, both set to 100 at the last 2025 close, through October 7, 2026. Sources: Taiwan Stock Exchange; Korea Exchange closing data.

The Market's Explanation

Investors offer a ready explanation. M&G Investments portfolio manager Vikas Pershad put it plainly: Taiwan's earnings "come from volume," while Korea's, "in the near-term, are coming from price." A sharper version is that markets are repricing scarcity.

Korea's core strength, memory, has historically been cyclical: capacity additions eventually discipline prices. Taiwan's strength sits in an ecosystem anchored by leading-edge logic, contract foundry and advanced packaging, where long qualification and limited capacity create longer-lived bottlenecks. In this telling, the semiconductor cycle has not died. The location and expected duration of the constraint have changed.

That hypothesis is attractive, which is exactly why it should be tested against three alternative explanations and valuation evidence before it is accepted.

Alternative Explanation 1: Concentration

Perhaps Taiwan's outperformance is just Taiwan Semiconductor Manufacturing Company (TSMC). The company carried a 41.12 percent weight in TAIEX on September 30, according to the Taiwan Futures Exchange. If TSMC were the story, the rest of the index would lag. It does not.

• TSMC shares rose from NT$1,550 at the end of 2025 to NT$2,480 on September 30, a 60.0 percent gain, while the index rose 65.5 percent over the same period.
• Using the September 30 weight as a simplifying proxy, IndraStra Global estimates that TSMC's return would account for roughly two-fifths of the index's gain on a simplified basis, implying a substantially stronger gain for the remainder of the index.
• About 10 percent of TAIEX stocks had at least doubled this year, against 4.1 percent of KOSPI stocks.

In Korea, the pattern runs the other way. Samsung Electronics (common and preferred shares) and SK hynix together averaged 52.82 percent of KOSPI in September, according to the Korea Financial Investment Association (KOFIA). On those weights, the calculations suggest that the two chipmakers' September weights and year-to-date returns imply a contribution of roughly four-fifths of the KOSPI's 62.3 percent gain through September 30, leaving a substantially smaller gain for the remainder of the index.

Note: Because the calculation uses September weights rather than daily constituent weights and does not perform formal index attribution, these figures should be treated as directional proxies rather than precise contribution measures.

Conclusion: The concentration story belongs to Korea, not Taiwan.

Alternative Explanation 2: Earnings Performance

Perhaps Taiwan earned more. It did not.

• Samsung reported second-quarter 2026 operating profit of ₩89.5 trillion, and SK hynix reported ₩60.54 trillion at a 76 percent operating margin.
• TSMC reported NT$766.6 billion at a 60.3 percent operating margin.
• Converted at the US Federal Reserve's H.10 second-quarter average rates (₩1,500.48 and NT$31.60 per US dollar):
• Samsung: US$59.7 billion
• SK hynix: US$40.4 billion
• TSMC: US$24.3 billion
• The two Korean companies added about US$90 billion of quarterly operating profit from a year earlier; TSMC added about US$9 billion.

The market's skepticism is therefore not about whether Korea can generate extraordinary semiconductor profits; it is about how much of those profits can survive when memory-price growth normalizes. That matters because the strongest cyclical pricing power in semiconductors currently sits in memory, particularly in AI-linked HBM and high-value DRAM.

Alternative Explanation 3: Valuation

Here the hypothesis survives in a modified form. If Korean profits are larger and growing faster, a 5.5-times multiple suggests that investors are heavily discounting their durability, rather than simply rewarding their current earnings.

• Earnings estimates for TAIEX rose about 19 percent in the September quarter against about 15 percent for KOSPI, and Taiwan's upgrades outpaced Korea's for the first time since March 2025.
• In the Bank of America fund-manager survey for September, 40 percent of respondents were overweight Taiwan against 25 percent for Korea.
• Asked which market would benefit most from the next phase of the AI cycle, 35 percent chose Taiwan and 5 percent chose Korea.
• Societe Generale strategists expect memory price appreciation to slow before normalizing in 2028 as Chinese competition rises.

Part of the valuation gap is the long-standing "Korea discount." A portion of the remaining gap can therefore be interpreted as a durability premium, although the Korea discount also reflects longer-standing structural factors: markets pay 18 times for profits they expect to persist and 5.5 times for profits they regard as closer to a cyclical peak.

Key Metrics Comparison

| Measure | Taiwan | South Korea | |---|---|---| | Benchmark change, 2026 to October 7 | TAIEX +72.0% | KOSPI +61.5% | | July–September 2026 | +3.9% | −19.3% | | Largest index weight(s) | TSMC 41.1% | Samsung + SK hynix 52.8% | | Approximate contribution proxy using cited weights (to September 30) | About 39% | About 81% | | Q2 2026 operating profit (US$) | TSMC US$24.3 billion | Samsung + SK hynix US$100.0 billion | | Increase from Q2 2025 (US$) | +US$9.3 billion | +US$90.1 billion | | One-year forward P/E (reported) | About 18× | About 5.5× |

Sources: Taiwan Stock Exchange, Korea Exchange closing data, Taiwan Futures Exchange, KOFIA, company filings, Federal Reserve H.10 (Q2 averages), reported market estimates; Contribution figures are IndraStra Global proxy estimates based on cited weights and constituent returns; they are not formal index-attribution calculations.

Why Markets Hesitate

The price data show why markets hesitate. TrendForce expects conventional DRAM contract prices to rise 10 to 15 percent in the fourth quarter. NAND flash prices are expected to rise 15 to 20 percent, and the market remains undersupplied. That follows a second-quarter DRAM surge of roughly 60 percent.

MOTIR's own fixed-price series shows a 16-gigabit DDR5 chip at US$48.00 in September, up from US$37.50 in May. That is deceleration, not collapse. In a highly cyclical market, however, investors often price the turn before it arrives, and memory makers know it. Long-term agreements with ceiling-price mechanisms, now spreading across the industry, trade some upside for predictability.

Why Taiwan's Constraints Should Last Longer

The answer lies in how an AI accelerator is built. A modern AI processor is not simply a single chip. It is a system: logic dies made on 3-nanometer or 2-nanometer processes, stacks of HBM (high-bandwidth memory) and a silicon interposer that joins them.

For many leading AI accelerators, that integration is performed through TSMC's CoWoS (chip-on-wafer-on-substrate) packaging, which TSMC describes as integrating multiple system-on-chip (SoC) dies with high-bandwidth memory (HBM) stacks. Each added layer raises complexity. Complexity can lengthen qualification, because a customer must prove a design on a specific process and packaging flow before committing volume. Long qualification cycles can concentrate orders in suppliers that have already passed. Concentration during a demand surge becomes a bottleneck. And a bottleneck confers pricing power for as long as it takes a rival to qualify.

Here "bottleneck" means control over a scarce production constraint, not a legal monopoly.

• TSMC held 73 percent of Counterpoint's pure-play foundry market in the second quarter, a tracker that counts Samsung's foundry business alongside dedicated foundries.
• In Counterpoint's broader "Foundry 2.0" measure, which adds non-memory chipmakers, outsourced assembly and test (OSAT) firms and photomask makers, its share was 42 percent.
• TrendForce described 3-nanometer supply as "a temporary single-supplier dynamic dominated by TSMC."
• It also noted that the CoWoS shortage running since 2023 now extends to equipment, substrates and materials.
• TSMC is qualifying a 5.5-reticle version of its CoWoS-L packaging this year, and its SoIC (System on Integrated Chips) 3D stacking entered volume production at 3 nanometers in 2025.

Supply-Chain Bottlenecks

| Supply-chain layer | Leading supplier(s) | Latest share or indicator | Diversification / erosion signal | |---|---|---|---| | Pure-play foundry (Q2 2026 revenue) | TSMC | 73%; Samsung Foundry 7% | Samsung 2nm wins; Rapidus 2nm plan | | CoWoS-class 2.5D packaging | TSMC-led ecosystem; ASE and other OSAT expansion | Supply-demand gap about 20% in 2026, about 10% in 2027 | ASE, Amkor expansion; Intel EMIB-T | | HBM (Q2 2026 revenue) | SK hynix, Samsung | 50%, 33%; Micron 18% | SK hynix down from 64% a year earlier | | Conventional DRAM (Q2 2026 revenue) | Samsung, SK hynix, Micron | 38%, 25%, 24% | ChangXin Memory Technologies (CXMT) 10%, up from 4% | | Extreme ultraviolet (EUV) lithography | ASML | Sole supplier | No rival EUV supplier | | T-glass fiber cloth for substrates | Nittobo | About 90% | New capacity mid-2027 at the earliest |

Sources: Counterpoint (foundry, Foundry 2.0, DRAM and HBM trackers), ASML, TrendForce. HBM shares may not sum to 100 because of rounding.

The HBM Complication

The comfortable version of this argument stops there, casting Korea as the cyclical memory supplier and Taiwan as the structural chokepoint. HBM breaks that picture.

HBM is itself a bottleneck, and Korea holds most of it. According to Counterpoint, SK hynix earned 50 percent of second-quarter HBM revenue and Samsung 33 percent, which means Korean suppliers accounted for roughly five-sixths of global HBM revenue. Korea therefore controls a major AI bottleneck of its own.

The East Asia Institute (EAI) identifies two bottlenecks in the AI supply chain: HBM and advanced foundries, one he...

[Content truncated]