← The Algorithm of Two Empires Vol. 2 17 / 18 한국어
Vol. 2 — The Algorithm of Two Empires

Chapter 16: The Investor's Map — A Positioning Framework for Hegemonic Transition

Opening: The Lesson of December 1989

On December 29, 1989, the Nikkei index hit 38,915 yen.

A major Tokyo brokerage published its year-end report under the title "Japan: Hegemon of the 21st Century." Even investors in Seoul or New York would not have dismissed it as hyperbole. That year, Japan accounted for 42% of the global equity market. A widely circulated calculation placed the estimated value of the Imperial Palace grounds in Tokyo above all real estate in California combined. The cyclically adjusted price-to-earnings ratio (CAPE) stood at roughly 100 times — double the peak of the American dot-com bubble.

Thirty-five years passed.

In 2024, the Nikkei barely recovered to that level. Over the same period, the S&P 500 rose fourteenfold on price alone, excluding dividends. An investor who went all-in on Japanese equities in December 1989 failed to break even in nominal terms for thirty-five years. Adjusted for inflation, the losses ran deeper. During those decades, 181 Japanese banks collapsed, and the phrase "lost thirty years" entered the textbooks.

Now the same question returns.

Is China's AI ascent a replay of Japan in 1989? Or is it the beginning of a genuine transition? And if the transition is real, how does one avoid repeating the mistake of the investor who went all-in on Japan?

The lesson of Japan is not simple. It is not "never invest in a challenger nation." Japanese corporations possessed real technological prowess. In automobiles, electronics, and semiconductors, Japan stood at the global frontier. The Nikkei's problem was not technology. It was price. A CAPE of 100 implied the assumption that growth would continue forever. When that assumption proved wrong, the reality of the technology did nothing to prevent the collapse of the price.

The deeper lesson is this: during a hegemonic transition, the greatest danger is going all-in on a single scenario.

Japanese investors in 1989 bet on one storyline: "Japan overtakes America." When that storyline failed, their entire portfolios crumbled. By contrast, investors who maintained global diversification and held American growth stocks alongside their Japanese positions turned those thirty-five years into opportunity.

In Chapter 15, we sketched three scenarios: the United States maintains technological hegemony, China overtakes, and the world fractures. This chapter asks the next question. How does an investor who cannot know which scenario will prevail position accordingly?

This chapter does not recommend specific securities. It presents a framework organized by scenario: what types of assets possess structurally favorable conditions under each outcome? A scenario-neutral core position, scenario-sensitive satellite positions, and adjustment rules triggered by transition signals. It offers separate lenses for individual and institutional investors.

Drawing a map and reading a map are different acts. This chapter is about how to read the map.

Investment Terminology for This Chapter

- CAPE (Cyclically Adjusted Price-to-Earnings Ratio): Share price divided by the average of ten years of inflation-adjusted earnings. The higher the number, the more expensive the stock. At 100 times, it means "recovering the current price would require a century's worth of earnings." The metric appears in this chapter as an overheating indicator for Japan's 1989 bubble.

- VIX: A volatility index derived from U.S. S&P 500 options markets. It quantifies how anxious investors are about the next thirty days, hence its nickname, the "fear index." Readings above 30 signal extreme market distress.

- CapEx (Capital Expenditure): Money a company invests in factories, data centers, equipment, and other assets for future growth. It is capitalized rather than immediately expensed. Big Tech spending on AI data center construction is a prime example.

- ETF (Exchange-Traded Fund): A fund that trades on a stock exchange like an individual share. Instead of picking single stocks, an investor can gain exposure to "all AI infrastructure" or "all Korean semiconductors" in one transaction. The simplest tool for diversification.

- Rebalancing: The process of restoring a portfolio to its original target weights after market movements cause drift. When one asset rises significantly, a portion is sold; proceeds are used to buy underweight assets, restoring the intended ratios.

- Hedge: A strategy of holding offsetting positions to protect against a specific risk. Buying gold is a hedge against dollar weakness or geopolitical crisis. Similar to insurance, but instead of paying premiums, one forgoes a portion of potential returns.

- Call option / Put option: Contracts that confer the right to buy (call) or sell (put) an asset at a specified price in the future. They allow asymmetric exposure to extreme scenarios with limited capital at risk. Put options that profit from a crash are a classic hedging instrument.

- Barbell strategy: Allocating to both high-risk and low-risk extremes while minimizing exposure to the middle. Like a barbell: heavy at both ends, empty in the center. A portfolio that combines aggressive AI positions with safe-haven assets is a typical example.

- P/E (Price-to-Earnings Ratio): Share price divided by earnings per share. It shows how much premium the market assigns to a company's profits. A P/E significantly above its historical average may signal overheating.

- Backtest: Applying an investment strategy to historical data to answer "what if we had done this back then?" A tool for stress-testing the logic of a strategy, though the past does not guarantee the future.

- CAGR (Compound Annual Growth Rate): An investment return expressed on an annualized basis. If an investment grew fivefold over thirty-five years, the CAGR is approximately 4.7%. Unlike a simple average, it accounts for compounding.

- VIE Structure: A legal structure Chinese companies use to circumvent foreign investment restrictions. Investors receive economic benefits, but legal ownership remains with a domestic Chinese entity. This is the source of a distinctive legal uncertainty in Chinese equity investment.

A minimum shared vocabulary for the framework that follows.

Section A. Historical Lessons of Investing Through Hegemonic Transitions

Rothschild's Carrier Pigeons

June 18, 1815. Waterloo.

Napoleon's army fell to Wellington. The first person in London to learn this outcome was not the British government. It was Nathan Rothschild. His network of carrier pigeons and fast boats delivered the news a full day ahead of official channels. That day, Rothschild bought British government bonds in volume. He acted before the market could process that the war was over — and therefore that British public finances were secure.

Two hundred and ten years have passed since then. The carrier pigeons of 2026 are Epoch AI benchmark data, IMF COFER reports, and hyperscaler quarterly earnings. Who reads first, who moves first. The structure has not changed.

The Rothschild episode endures in investment history not because of its returns, but because it was the first modern demonstration that information asymmetry is the fulcrum of asset allocation. And during hegemonic transitions, that asymmetry reaches its most extreme.

From Britain to America — A Thirty-Year Transition

The transfer of hegemony from Britain to the United States was history's only "peaceful hegemonic transition." Its timeline still holds lessons for investors today.

On the eve of World War I, Britain held 40% of the world's overseas investment. London was the center of global finance. The pound was the reserve currency. British government bonds (consols) were synonymous with the "risk-free asset."

War changed all of that.

By the mid-1920s, the dollar had overtaken the pound, first in trade credit, then in foreign exchange reserves. In 1931, the pound left the gold standard. Sterling holders bore the full cost of devaluation. In 1945, the Bretton Woods system formalized dollar hegemony. And through the 1970s, the pound lingered on for another three decades as a "zombie international currency."

Four lessons emerge from this process.

First, reserve currency transitions are gradual but irreversible. An investor who recognized the dollar's rise in the 1920s carried a thirty-year advantage. But the transition did not happen overnight. The shift from sterling to dollar unfolded over thirty to fifty years in stages.

Second, the hegemon's assets remain overvalued until just before the fall. British bonds were safe-haven assets right up to the eve of World War I. The signals that hegemony was eroding were present, but there was a lag before prices reflected them.

Third, the rising empire's infrastructure assets produced the highest returns. U.S. steel output surpassed Britain's in 1880 (2.5 million tons versus 2 million). By 1900, it led the world at 10 million tons. Capital invested in American railroads, electrical grids, and telecommunications infrastructure recorded the highest returns from the 1900s through the 1930s.

Fourth, diversification during the transition was the condition for survival. Capital concentrated in a single empire could not escape catastrophic losses when the transition came.

The Third Lesson of the Japanese Bubble

The story of Japan in 1989 yields more than just the first lesson.

A CAPE of 100 was a warning in itself. But investors in Tokyo at the time were not buying a ratio; they were buying a narrative. The "Japan Model" narrative. A story in which the state designs industry, corporations are bound by long-term relationships, and workers are secured by lifetime employment — and this model defeats the American free market. When a narrative is powerful enough, the numbers become invisible.

The second lesson is the absolute importance of diversification. The third lesson is less well known but more important.

Japan's technological capabilities were real. Toyota's production system remains the textbook of world manufacturing to this day. Sony, Panasonic, and Sharp dominated global electronics markets in the 1980s. NEC, Fujitsu, and Hitachi competed head-to-head with Intel in semiconductors. Technology and asset prices are separate things. Even when the technology is real, if its value has already been excessively priced in, investment returns will not follow.

The same lens must be applied to China's AI capabilities. DeepSeek's efficiency breakthrough is real. The energy advantage is real. But reality does not automatically translate into investment returns. The distance between technology and price is what matters.

Present Coordinates

As analyzed in Chapter 15, the dollar's share of global reserves is declining, but the renminbi (at 2%) remains far from a viable alternative. Abandoning the dollar in the short term would be the same error as abandoning sterling too early in the 1920s.

AI infrastructure is today's "railroads and electrical grids." BlackRock projects that cumulative AI-related capital expenditure through 2030 will reach $5 trillion to $8 trillion. That demand does not disappear under any scenario.

China's real estate crisis bears structural resemblance to Japan's bubble collapse. Chinese property prices have fallen more than 40% from their peak. One hopes the People's Bank of China will not repeat the Bank of Japan's mistakes of 1989 — but history has seldom rewarded optimism.

Section B. Scenario-Based Asset Allocation Framework

An Afternoon on Yeouido

Yeouido, Seoul. The conference room of an asset management firm. Three o'clock in the afternoon.

Two fund managers stand before a whiteboard. On the left: "Portfolio A." On the right: "Portfolio B."

Portfolio A is a concentrated U.S. AI play. NVIDIA, Microsoft, Amazon, and Google together account for 60% of holdings. The annualized return over the past three years: 34%.

Portfolio B is a U.S.-China diversified strategy. U.S. AI 30%, Korean semiconductors 15%, energy 15%, emerging Asia 15%, bonds and gold 25%. Over the same period: 22% annualized.

On the numbers alone, A dominates. But then the next question surfaces: "Will A still be right next year?" No one can say with certainty. If Scenario A (U.S. maintains dominance) continues, A is correct. If the world shifts to Scenario C (fracture), B survives. And if you cannot know the future, you must design your position without knowing it.

This is the starting point of the framework presented in this section.

The Three-Layer Portfolio

A portfolio is constructed in three layers.

Layer 1: Scenario-Neutral Core (Core Neutral, 30–40%) Assets whose demand persists under every scenario. Whether the United States wins, China wins, or the world splits apart, these are needed regardless.

Asset Weight Rationale
AI Infrastructure (U.S. + Global) 15% Hyperscaler CapEx $660–690B — scenario-independent
Energy (Nuclear + Renewables) 10% Data center power demand 800–1,050 TWh in 2026 (IEA base to upper) — 2–3x growth
Korean HBM & Semiconductors 5% "The shovel's shovel" — supplying both sides
Gold 5% Geopolitical risk hedge
Short-term government bonds 5% Liquidity and optionality

This 40% is not a "scenario bet." It is "scenario insurance." Whichever scenario materializes, this core holds.

Layer 2: Scenario-Sensitive Satellites (30–40%) Positions to expand as a given scenario strengthens.

Satellites for Scenario A (U.S. maintains dominance): Additional 15% in U.S. AI infrastructure (Big Tech, GPUs, cloud). Additional 10% in U.S. growth stocks (AI-native). Additional 5% in U.S. energy (nuclear, renewables). 5% U.S. Treasuries. 5% selective China exposure (consumer, AI applications).

Satellites for Scenario B (China overtakes): 12% in Chinese AI applications (ByteDance ecosystem, AI SaaS). 10% in the Global South (India, Southeast Asia, Middle East). 7% in energy and commodities. 6% in U.S. defensive assets (dividends, healthcare). 5% in gold and commodities.

Satellites for Scenario C (fractured world): 10% in AI infrastructure diversified across both blocs. 10% in "dual supply chain" beneficiaries (Korean semiconductors, Japanese materials, German equipment). 5% in defense and cybersecurity. 8% in diversified energy. 7% in gold and real assets.

Layer 3: Options and Hedges (5–10%) Protection against extreme scenarios. A Taiwan Strait crisis, an AI bubble crash, premature AGI achievement. Asymmetric exposure to tail events.

Scenario-Weighted Positioning

An institutional investor can assign weights to the three scenarios. This book's model portfolio weights Scenario A at 50%, Scenario C at 30%, and Scenario B at 20%. These ratios reflect the author's analytical framework, a subjective judgment grounded in the current positions of the six variables analyzed in Chapter 15.

These weights are not fixed. They must be recalibrated whenever the direction of the ten signals below shifts. If three or more signals favoring Scenario A reverse simultaneously, the principle is to reduce A's weight and raise C or B accordingly.

In BlackRock's 2026 survey, 54% of respondents said they see better AI-related opportunities in energy companies than in Big Tech. This is a market signal to increase the energy weight within the scenario-neutral core. Under every scenario, data centers need power.

An analysis of 901 advisor portfolios found that tech weightings averaged 9% below the S&P 500 benchmark, another signal. Either the market has not fully priced in the AI theme, or it is deliberately avoiding excessive concentration. One or the other.

Execution Example: Scenario-Weighted Model Portfolio

In concrete terms, here is what a portfolio looks like when weighted at Scenario A 50%, Scenario C 30%, Scenario B 20%.

Layer 1 — Scenario-Neutral Core (40%) Global AI infrastructure ETF 15%. Global energy (conventional + renewables) 10%. U.S. mid-cap growth 5%. Asia ex-China 5%. Global bonds (short + intermediate) 5%.

Layer 2 — Weighted Satellites (50%) Scenario A satellites × 50% = 25% allocation: U.S. Big Tech + AI leaders 8%, U.S. defense and cybersecurity 5%, dollar-denominated asset reinforcement 5%, allied-nation technology (Japan, Korea, Taiwan) 4%, emerging-market digital transformation 3%. Scenario C satellites × 30% = 15% allocation: AI infrastructure diversified across blocs 4%, dual supply chain beneficiaries (Korean semiconductors, German equipment) 4%, diversified energy 3%, gold and real assets 2%, defense and cybersecurity 2%. Scenario B satellites × 20% = 10% allocation: Chinese AI applications 3%, Global South 3%, energy and commodities 2%, gold 1%, U.S. defensive assets 1%.

Layer 3 — Options and Hedges (10%) VIX call options 3%. Physical gold and gold ETF 3%. Taiwan Strait risk put options 2%. Long-duration government bonds (20-year+) 2%.

Total: 100%. This is a starting point, not an optimum. What matters is the structure: under any scenario, the portfolio's 40% core holds the neutral position, and the remaining 50% is redistributed as scenario weights shift.

Rebalancing Triggers

Maintaining a portfolio is harder than building one. In addition to regular rebalancing (once per quarter), an immediate review is triggered if any of the following conditions are met.

Scenario transition trigger: When three or more of the ten signals listed above point simultaneously in the same direction. If, for example, a benchmark reversal, a decline in the dollar's reserve share, and an ally's defection all appear at once, reduce Scenario A's weight and raise B or C.

Volatility trigger: When the VIX exceeds 30, or when any satellite asset class falls more than 20% within a single quarter. In such cases, reduce satellite exposure and temporarily increase the weight of Layer 1 (neutral core) and Layer 3 (hedges).

Valuation trigger: When a specific sector's P/E exceeds twice its ten-year average, or when earnings revisions trend downward for three consecutive months. During the dot-com bubble in 2000, the Nasdaq's P/E exceeded three times its ten-year average. Those who read the signal survived.

Currency Hedging

In a portfolio built for the era of U.S.-China competition, currency risk is as large as asset risk.

Base rule: When non-reserve-currency-denominated assets (renminbi, won, yen, etc.) exceed 25% of the portfolio, hedge 50% of the excess through currency forwards or currency ETFs. Dollar-denominated assets serve as a natural hedge under Scenarios A and B, making separate hedging unnecessary. Only under Scenario C (fracture) does the possibility of dollar weakness itself open up — and in that case, gold serves as a complementary currency hedge.

If this rule seems simple, that is by design. Complex hedging strategies generate execution costs that erode expected returns. The principle is singular: check every quarter whether you are excessively exposed to any single currency, and trim only the excess.

Section C. Sector Positioning — Finding the New Railroads

SEMICON Korea 2026

January 2026. COEX, Seoul. SEMICON Korea.

The most frequently heard phrase at the exhibition entrance: "Physical AI." Two years earlier, in the same venue, the dominant phrase was "generative AI." An analyst scribbles in a notepad: "The next wave of digital AI is the physical world. That wave will need memory too."

Lines form in front of the Samsung Electronics and SK hynix booths. The people waiting are procurement officers from NVIDIA, Google, and Microsoft. What they have come to see: HBM4 samples. Memory for next-generation AI accelerators. Without this small chip, their data centers stop.

AI Infrastructure — The New Railroads

What was the surest investment opportunity in nineteenth-century America? Not the Gold Rush itself. The railroads. When no one knew which prospector would strike gold, the company that laid the rail collected fees no matter who won.

The railroads of 2026 are AI infrastructure.

Combined CapEx for the five major hyperscalers is projected at $660 billion to $690 billion. On a single-year basis, that figure rivals the total revenue of the entire global semiconductor industry. Half of this spending goes toward inference workloads. This represents a structural shift in the investment thesis: from training to deployment, from research to service. It marks the moment AI transitions from "experiment" to "infrastructure."

Between 2026 and 2030, data center capacity will grow by approximately 100 GW, an asset value of $1.2 trillion. The components that make up this infrastructure (compute, networking, cooling, power) are today's "ties and rails."

Semiconductors — The Physical Foundation of Hegemony

AI is software. But software does not exist without hardware. And the core of the hardware is the semiconductor.

NVIDIA's GPU market share stands at 92%. In the AI chip market, it holds 90%; in AI data center revenue, 86%. Its FY2027 backlog is $320 billion. This number does not mean "demand exists"; it means "demand is already locked in."

Yet something deserves even more attention than GPUs: HBM, High Bandwidth Memory.

No matter how fast a GPU computes, if the memory feeding it data is slow, the GPU waits. Memory bandwidth in AI workloads operates as a bottleneck in ways fundamentally different from the CPU era. A single GPU requires eight to twelve HBM units. One NVIDIA Blackwell chip contains eight HBM3E modules.

The global memory semiconductor market is projected to grow 98% year-over-year in 2026, reaching $445 billion. Korean companies (Samsung Electronics and SK hynix) command the bulk of this market. SK hynix holds 57–60% of HBM market share. HBM3E prices have risen 20% on 2026 delivery contracts. Across server DRAM as a whole, price increases of up to 70% are being demanded.

In the Gold Rush, the most consistent profits went not to those who panned for gold but to those who sold shovels. The shovel of the AI era is the GPU. And Korean HBM is "the shovel used to make the shovel." Just as AI computation is impossible without NVIDIA, NVIDIA cannot deliver its full performance without HBM.

ASML's EUV lithography equipment occupies the same structural position. It is the only tool capable of manufacturing advanced semiconductors. Only one company in the world makes it, and that company has not sold a single unit to China. This is "the shovel's shovel's shovel."

Energy — The Hidden Bottleneck

AI consumes electricity.

Global data center power consumption in 2026 stands at approximately 800 TWh under the IEA's base projection and 1,050 TWh under its upper scenario, more than double the 2022 level. The United States will see 240 TWh of additional demand by 2030. China will add 175 TWh. Together, they account for 80% of the global increase in data center power consumption.

This is why Big Tech companies are racing to announce small modular reactor (SMR) plans. The combined SMR commitments of Microsoft, Google, and Amazon exceed 20 GW. But SMR commercialization will arrive no earlier than 2030 (realistically, 2032 to 2035). In the interim, the power gap must be filled by restarting existing nuclear plants and expanding renewables.

Bottlenecks always hide where no one looks. When people think of AI investment, they think of NVIDIA or OpenAI, not power utilities. But the market is already reading this structure. The BlackRock survey cited in Section B is the evidence.

When the Mississippi Bubble and South Sea Bubble burst in 1720, the most stable returns went to the bill clearinghouses. Whenever the AI bubble arrives, power infrastructure will survive it.

Robotics — The Next Wave

"Physical AI" at SEMICON Korea 2026 is not mere marketing jargon.

Tesla targets production of 50,000 Optimus robots in 2026. The projected total market for humanoid robots by 2050: $5 trillion, larger than the automotive industry.

Why now? Because China has structural demand.

As analyzed in Chapter 15, China's demographic cliff (a total fertility rate of 1.0, with the share of the population aged 60 and above set to breach 30% by 2035) creates a structural imperative to compensate for labor shortfalls with robots. This is not a strategic choice. It is an unavoidable demand.

That demand connects directly to the Korean memory industry. Humanoid robots require high-performance memory for onboard AI computation. "The next wave of Physical AI will need memory too." That analyst's note was precisely correct.

Section D. Scenario Transition Signals and Monitoring Framework

A Hypothetical Morning in 2028

One morning in 2028 (hypothetical): Epoch AI releases its quarterly report.

For the first time, a Chinese frontier model has surpassed an American model on major benchmarks. The margin is small — 0.5 percentage points. But the direction has reversed. That same day, the NASDAQ falls 4%. The Shanghai Composite rises 3%. Social media erupts: "AI hegemony has shifted."

The investor who prepared for this signal in advance and the investor who encountered it for the first time that morning reacted differently.

The prepared investor did not panic. She recognized that the conditions for expanding her Scenario B satellite position had been met. She executed the rebalancing protocol she had defined in advance. No panic selling, no impulsive buying.

The investor without a signal framework spent the day watching the NASDAQ fall, searching for answers that existed nowhere.

Ten Core Signals

During a hegemonic transition, scenarios are not predicted — they are monitored. Track the following ten signals, and you can discern which scenario is gaining strength.

Signals indicating Scenario A is weakening:

Signals indicating Scenario B is strengthening:

Signals indicating Scenario C is strengthening:

Monitoring Cadence

A tracking cadence is established for these signals.

Quarterly (3 months): Epoch AI benchmark data. IMF COFER dollar share. Hyperscaler CapEx actuals. NVIDIA backlog updates.

Semi-annually (6 months): Taiwan Strait security situation. Chinese GDP growth rate. SMIC and Huawei Ascend production progress. SK hynix and Samsung HBM market share.

Annually (12 months): WEF Global Risks Report. BlackRock AI risk dashboard. U.S. federal AI legislative status. Progress in global AI standards negotiations.

This monitoring does not automate investment decisions. But it makes response possible instead of reaction.

Risk Matrix

Three extreme risks are managed separately.

Taiwan Strait crisis: Low probability, extreme severity. If it occurs: TSMC supply cut → global semiconductor supply chain collapse → rapid shift to Scenario C. The hedge consists of geographic diversification (reducing Taiwan semiconductor dependency) and defense/cybersecurity exposure.

AI bubble correction: Medium probability, medium severity. Triggered if AI revenue monetization lags hyperscaler CapEx. A pattern analogous to the dot-com bubble of 2000. The hedge consists of energy, gold, and physical infrastructure.

Premature AGI achievement: Low probability, extreme severity. If a U.S. company achieves it first, Scenario A becomes irreversibly entrenched. In that case, rapidly reduce Scenario B and C satellite positions and maximize Scenario A satellites. If this signal fires, all other monitoring rules are reset.

Section E. Different Lenses for Individual and Institutional Investors

Two Portfolios, Thirty-Five Years Later

December 1989. Two investors.

Investor A went all-in on Japanese equities. He read the "Japan: Hegemon of the 21st Century" report and believed it. At Nikkei 38,915, he filled 100% of his portfolio with Japanese stocks.

Investor B made a different choice. Japanese equities 30%, U.S. growth stocks 30%, European diversification 20%, bonds 20%. It was not that Investor B disbelieved Japan's ascent. He simply acknowledged the possibility of being wrong. So he diversified.

The results thirty-five years later marked a divergence.

Investor A recovered his nominal principal only in 2024. His real return, adjusted for inflation, was negative. Investor B's portfolio grew more than fivefold over the same period. Investor B was not smarter. He simply designed his portfolio to accommodate the possibility of error.

Historical Validation: 1989–2024 Backtest

Compare the outcomes of three investors who committed equal amounts in late December 1989.

| Investor | Strategy | 35-Year Cumulative Return | Annualized (CAGR) |

|—————|—————|—————————————|—————————-|

| Investor A | Nikkei 225, 100% | +54%\* | ~1.2% |

| Investor B | Japan 30% + U.S. 30% + Europe 20% + Bonds 20% | ~500%+ | ~5–6% |

| Investor C | Global 60/40 (MSCI World 60% + Global Bonds 40%) | ~900%+ | ~7–8% |

\* Total return in yen terms, including dividend reinvestment. On price return alone, nominal 0% (breakeven). In dollar terms, +42% (source: CME Group, 2024).

Over the same period, S&P 500 total return (with dividends reinvested) averaged 10.7% annualized — 32 times cumulative over thirty-five years (source: Damodaran, NYU Stern, 2024). The global 60/40 portfolio returned 8.2% annualized over thirty years (source: LazyPortfolioETF; Dimson-Marsh-Staunton, Triumph of the Optimists).

Core lesson: Diversification works not because it maximizes returns, but because it prevents a single scenario's failure from destroying the whole. Investor A avoided "worse" assets for thirty-five years, yet that very choice produced paradoxically the worst outcome. Investor C survived not because she avoided betting on Japan, but because she avoided betting on any single country.

Implications for 2026: Going 100% into U.S. AI today carries the same structural risk as going 100% into Japan in 1989. The reality of AI technology does not guarantee the rationality of its price. That is the single message this backtest delivers.

This is the core of the framework for individual investors.

Individual Investors: A Structure That Prepares for Scenarios

For individual investors, correctly predicting which scenario will materialize is close to impossible. Warren Buffett cannot do it. Ray Dalio cannot do it. No individual can.

What is possible, instead, is building a structure that can endure any scenario.

Recommended structure:

Scenario-neutral core, 50%: Global AI infrastructure ETF, energy ETF, gold. This position requires only one thing to be true: that AI continues.

U.S. AI growth, 20%: S&P 500 or NASDAQ-based index. Valid as long as Scenario A persists.

Asian semiconductors, 10%: Korean semiconductor ETF or individual companies. Close to scenario-neutral, but classified separately due to Korea-specific geopolitical risk.

Global South, 10%: India, Indonesia, and Brazil ETFs. A "non-aligned" position that benefits under both Scenario B and Scenario C.

Options and hedges, 10%: Additional gold, short-term government bonds. Liquidity reserves.

The key to this structure is not "maximum returns if the U.S. wins" but "losses within an acceptable range even if the U.S. loses."

Four Traps for Individual Investors

The four most common traps individual investors fall into during a hegemonic transition.

Trap 1: Single-country all-in. Betting on one scenario: "America will certainly win" or "China will certainly rise." This is the trap Investor A fell into in 1989. Certainty is the most expensive conviction in a transition.

Trap 2: AI bubble panic selling. Liquidating all AI-related assets during a correction. Investors who sold every internet stock during the dot-com bubble of 1999 missed Amazon and Google from 2003 onward. A correction and a collapse are different things.

Trap 3: Complete avoidance of Chinese assets. VIE structure risk, capital controls, and regulatory risk are real. But completely excluding the world's second-largest economy carries significant opportunity cost. "Selective exposure" is more rational than "total avoidance."

Trap 4: Overreacting to short-term news. Restructuring an entire portfolio on the basis of a single Epoch AI report or a single IMF forecast revision. The ability to distinguish signal from noise is essential. This is precisely the purpose of the monitoring cadence defined above.

Section F. Connecting to Volume 1 — Extending "The Shovel Seller"

Why the Gold Rush Lesson Endures

In the California Gold Rush of 1848, the people who got rich were not the ones who panned for gold.

Among the prospectors, few ended up wealthy. Most dug and went home. Levi Strauss, on the other hand, sold denim trousers and lasted. The merchants who sold shovels profited regardless of where the gold turned up. This was the essence of "the shovel seller."

In the AI-era Gold Rush, what is the shovel?

First-order shovel: NVIDIA GPUs. The foundational tool of AI computation. Regardless of which AI company prevails, training and inference require GPUs. (Market share and backlog figures appear in the preceding section.)

Second-order shovel: HBM from SK hynix and Samsung. The condition for the GPU to perform at full capacity. "The shovel used to make the shovel." Whether American AI or Chinese AI wins, HBM is needed.

Third-order shovel: ASML's EUV lithography equipment. The only tool that can manufacture HBM and GPUs. Only one company in the world makes it. "The shovel's shovel's shovel."

Energy infrastructure. The power that drives AI computation. Every Gold Rush needed water. The AI Gold Rush needs electricity.

In Volume 1, "the shovel seller" was a career strategy for individuals. Now we extend that concept into an investment framework.

The Dematerialization of Leverage — Once More

Volume 1 tracked a pattern: the source of power becomes progressively less material. From land to factories, from factories to API calls.

The Gold Rush shovel was a physical tool. The AI-era shovel is half-dematerialized. The NVIDIA GPU is physical, but the core of its value lies in the CUDA architecture, a software ecosystem. Thousands of libraries, hundreds of thousands of engineers, and two decades of accumulated knowledge constitute that value. Replicating this takes far longer than replicating the physical manufacturing.

The value of SK hynix's HBM follows the same logic. 3D stacking technology, close collaboration with NVIDIA, yield management know-how. This is not a factory; it is a design capability. There are reasons China is attempting HBM self-sufficiency, and there are reasons it remains difficult. A physical fab can be built. A decade of know-how cannot be copied.

But the dematerialization of leverage cuts both ways. Physical fabs are not easily replicated, yet the software layer can be copied far more quickly. When DeepSeek released R1, American competitors absorbed the same techniques within days. In the world of open source, a "shovel" commoditizes in months.

The "shovel seller" of the AI era is whoever holds the non-replicable layer. NVIDIA's CUDA ecosystem, SK hynix's HBM know-how, ASML's EUV monopoly. These are the shovels of this moment.

The Expiration Date of the Shovel

But even shovels have an expiration date.

SMIC is producing 60,000 wafers per month without EUV in 2026. By 2027, the target is 80,000. Huawei's Ascend 910C will produce 1.6 million dies in 2026. Performance-to-cost ratios trail NVIDIA, but the gap is narrowing.

The same applies to Korean memory. CXMT and other Chinese firms are pursuing HBM self-sufficiency. The timing of success is uncertain, but the attempt is already underway.

For investors, the implication is clear: do not depend forever on "today's shovel." Realize gains while the shovel's value is at its peak, and begin searching for the next shovel simultaneously.

In Volume 1, Arkwright's water frame received its patent in 1769. The steam-powered spinning machine arrived in 1785. The water frame's shelf life was sixteen years. The shovels of the AI era may be replaced even faster.

Transition: The Map Is Complete, But —

Investing during a hegemonic transition is not a game of prediction. It is a game of preparation.

No one precisely predicted the collapse of the Japanese bubble in 1989. But the investors who maintained diversification, monitored the signals, and adjusted their positions when the scent of a bubble grew strong — they survived.

This chapter's framework consists of three elements.

First, always maintain a scenario-neutral core. Assets whose demand does not vanish under any scenario — AI infrastructure, energy, the shovel sellers.

Second, adjust satellite positions according to signals. Ten core signals and a quarterly-semi-annual-annual review cadence.

Third, never go all-in on a single scenario. The most expensive lesson that Tokyo in 1989 ever taught.

Yet something is missing from this map.

NVIDIA designs. TSMC manufactures. ASML monopolizes the equipment. SK hynix supplies the memory. Neither the United States nor China possesses a self-contained supply chain on its own. The nations that hold the critical nodes of this supply chain share a single strategy — to become indispensable in a world where two empires' algorithms are running.

In SK hynix's M16 fab in Icheon, Gyeonggi Province, HBM4 chips are being produced at this very moment. Those chips ship to San Francisco, to Shenzhen, to Abu Dhabi. A single factory in a single country feeds algorithms across three continents.

Leverage has an expiration date. The question is what you do before it expires.

[Volume 1 Connection]

The core concept of this chapter — "the shovel seller" — first appeared in Volume 1, Chapter 17. There, the Gold Rush served as a historical case study for individual career strategy. This chapter applies the concept against the macro backdrop of U.S.-China hegemonic competition, extending it into concrete corporate archetypes: NVIDIA, SK hynix, and ASML.

Volume 1's "dematerialization of leverage" is reinterpreted here through an investment lens. The source of power that migrated from land to factories, from factories to API calls, now exists in a semi-materialized form — GPU design ecosystems and HBM know-how. This is the essence of today's "shovel."

Volume 1's "Engels' Pause" reappears in this chapter as an investment risk. An AI investment structure in which productivity rises but monetization lags — this is the essence of the AI bubble risk. During the early decades of the Industrial Revolution, a structural gap opened between productivity and wages. The gap between AI capital expenditure and AI revenue may be its modern counterpart.

Citations and references for this chapter can be found in the Vol. 2 bibliography.