The Korean Central Bank's Wager: Is the AI Supercycle a Structural Turning Point or a High-Stakes Gamble?

The Event

In July 2026, the Korean Central Bank released a report dismissing investor concerns that the semiconductor supercycle has peaked. The bank claimed the global semiconductor market remains in a supply shortage state, with the chip supercycle driven by artificial intelligence expected to persist for some time. The core argument: this chip cycle differs from the past—its driver is not traditional inventory cycles or capacity utilization fluctuations, but rather major enterprises competing to invest based on expectations that AI proliferation will trigger fundamental ecosystem transformation across industries.

Observation: The Trap of "This Time Is Different"

The semiconductor industry indeed faces two competing forces:

Cyclical signals: Between 2024-2025, chip spot prices retreated from highs, inventory cycles began adjusting, and some second-tier foundry orders declined. Many investors interpreted this as a "supercycle peak" signal—a pattern the semiconductor industry has experienced every 3-5 years over the past 30 years.

Structural signals: Simultaneously, NVIDIA and AMD experienced explosive demand for AI server chips, with order booking cycles extending to 2027-2028. Major cloud computing companies (OpenAI's backers, Google, Meta, etc.) announced multi-billion-dollar AI infrastructure investment plans based on long-term forecasts of LLM training and inference computing demand, rather than short-term earnings considerations.

The Korean central bank's argument is: this time, structural demand is large enough to absorb the impact of cyclical fluctuations.

History's Lessons

But "this time is different" are the seven most dangerous words on Wall Street.

Looking back at semiconductor history:

• 1990s: PC proliferation was viewed as a "permanent turning point," and memory chip investment went wild. When the internet bubble burst in 1998-2001, chip inventory piled up like mountains, and prices were cut in half.
• 2008-2010: Smartphones were seen as a "permanent turning point," and Qualcomm and Samsung's investments exploded. In 2011-2012, the smartphone market saturated, power amplifiers and baseband chips became oversupplied, and some suppliers went bankrupt.
• 2016-2018: Cryptocurrency mining was viewed as a "permanent turning point," and GPU manufacturers made fortunes daily. During the 2018 crypto winter, the GPU market collapsed, and NVIDIA's stock price fell more than 60% in 18 months.

Each time, boundaries were mismeasured. Not because analysts were incompetent, but because "structural turning point" itself is filled with measurement error.

The True Demand Curve for AI

Whether the Korean central bank's argument holds depends on three propositions:

Proposition 1: Does AI training compute possess an "irreversible cost curve decline"?

Historical patterns show that once chip costs decline, users find more lower-end, more marginal applications. This flattens the slope of the demand curve and puts pressure on prices. If DeepSeek's inference costs can truly drop to 1/25 of OpenAI's, it will spark hundreds of times more marginal use cases—but will there be enough of them? Will they be sticky enough? Hard to say.

Proposition 2: Does corporate investment have a "self-fulfilling expectation" loop?

Part of the AI investment announced by major cloud computing companies will indeed materialize, but some will be scaled back or delayed because of: (a) actual ROI falling short of expectations, (b) competitors' excessive investment causing average returns to decline, (c) regulatory policy changes. Every moment in history when "everyone is betting on X," the result is some players push chips in only to discover someone at the table has already gone all-in.

Proposition 3: Is the supply chain truly "undersupplied," or is it "high-end products undersupplied, low-end products oversupplied"?

The 2024-2025 phenomenon is: high-end AI chips (H100, H200, next-gen GPUs) have order lead times extending to 2027, but mid-to-low-end chips (older-generation GPUs, CPUs, edge computing chips) are accumulating inventory. This is not "global undersupply," but rather "high-end undersupply, low-end oversupply." The central bank's phrasing easily obscures this detail.

Practical Methods to Distinguish Structural from Cyclical

In real time, it's hard to distinguish between the two. But three indicators can help approximate the judgment:

1. Cost curve slope: If marginal costs continue declining (e.g., 7nm → 5nm → 3nm process energy efficiency), structural demand exists; if the cost curve has flattened (Moore's Law slowing), cyclical signals dominate.

2. Investment annualized rate vs. cash flow returns: If manufacturers' investment payback periods exceed 7 years and investment amounts are still increasing, the investment logic is mostly "belief in long-term structural change"; if payback periods are 3-4 years yet investment amounts are declining, it signals cycle peak warning.

3. Secondary market pricing: If the market assigns AI chip companies price-to-earnings (P/E) ratios with more than 50% premium relative to historical average, and the premium is widening, the market believes in structural turning point; if the premium is narrowing, the market is beginning to doubt.

The Korean Central Bank's True Intent

Notably, the Korean central bank's report was released in July 2026, precisely when the market began doubting whether the chip cycle has peaked. South Korea is the world's second-largest chip manufacturing nation (Samsung) and a major supplier of materials and equipment (SK Hynix, SK Innovation, etc.). Maintaining optimistic expectations about the semiconductor market aligns with national economic interests. The central bank's report likely carries dual meanings of both "economic forecast" and "policy signal"—calming markets, stabilizing chip manufacturers' investment confidence, thereby stabilizing exports.

This is not to say the report is wrong, but rather: instead of trusting the central bank's conclusions, it's better to treat the central bank report itself as a "market psychology signal"—when central banks feel compelled to clarify, it often signals the market is already doubting.

Historical Analogies

In 2007, Federal Reserve Chair Ben Bernanke said "subprime mortgage risks are under control"; in 2021, the Fed said "inflation is only temporary"; in 2023, central banks said "the banking system is stable"—then everything exploded three months later.

Central banks are the institutions with the most information, but also the ones most resistant to accepting that "the sectors we've bet on might be peaking."