The AI Infrastructure & Semiconductor Investment Atlas

Global Systems, Supply Chains, and Security Analysis

Reader orientation

This atlas is a map of the global AI economy and an investment decision framework. It distinguishes three questions that are often blurred together:

1. Is a company strategically important to the AI stack?
2. Does that importance materially affect the security an investor can own?
3. Does the current price offer an attractive expected return?

The first question determines coverage. The second determines investability. The third determines action. A monopoly, national champion, or private frontier lab can be essential to the map without being an attractive or directly investable security.

How to use the atlas

• For the thesis: read Part I, then use Part IV to see the scenarios and monitoring framework.
• For infrastructure and supply chains: read §2.3 and §§2.5–2.13, which follow the system from chip and rack to a commissioned data center.
• For security analysis: start with Part V, then return to the relevant stack chapter to test the operating thesis.

Contents

1. Part I — The Argument
2. Part II — The Stack, Layer by Layer
3. Chapter 2.13 — From GPU to AI Factory
4. Part III — Systems in the Round
5. Part IV — Where It Is Going
6. Part V — Investment Decision Layer
7. Part VI — Reference and Apparatus

Ten conclusions

1. AI demand is real, but infrastructure revenue still outruns disclosed end-customer economics. The durability question is not whether models improve; it is whether production workloads produce enough cash to service the capital built for them.
2. The bottleneck migrates. Accelerator scarcity moved into HBM, packaging, networking, power, and construction. A durable investment process follows lead times and capacity response rather than yesterday's shortage.
3. Strategic importance does not equal equity capture. Parent dilution, regulated pricing, customer bargaining power, state ownership, and valuation determine how much supply-chain leverage reaches shareholders.
4. Layer diversification can conceal one common factor. Much of the Western stack depends on a small group of hyperscaler capital budgets; much of the Chinese stack depends on state procurement and domestic fabrication capacity.
5. Inference expands the addressable market and intensifies the depreciation debate. Falling unit costs support usage, while rapid hardware turnover can weaken reported returns even when demand remains strong.
6. The allied chokepoint system is as important as either superpower. Taiwan, Japan, Korea, and the Netherlands control fabrication, memory, lithography, substrates, and engineered materials that neither the United States nor China can quickly replace.
7. The third bloc is a demand system, not a neutral spectator. Gulf sovereign capital, Indian and Southeast Asian capacity, European regulation, and non-aligned procurement influence which technical stack reaches global scale.
8. Power is regional. National electricity forecasts do not determine whether a specific campus can secure land, interconnection, equipment, water, permits, labor, and an economically acceptable tariff.
9. Private-company marks and public-market prices are different objects. Private strategic importance can be worth tracking even when no liquid security exists; the actionable shortlist therefore focuses on securities with observable prices and scenario-based valuation discipline.
10. The atlas is a dated decision system, not a timeless ranking. Prices, sanctions, private valuations, model leadership, and quarterly spending require scheduled refreshes and explicit stale-after dates.

Confidence language

The manuscript uses the following reader-facing states:

• Observed: reported historical or current data.
• Company-reported: an issuer or private company metric that is not independently audited for the purpose used.
• Estimate: a transparent third-party or modeled value.
• Forecast: a forward projection attributed to its source.
• Analytical synthesis: a conclusion or scenario constructed from multiple inputs.
• Contested: credible sources materially disagree or the available evidence is weak.

Market prices in the investment decision layer are observations at the latest available close before the cutoff. Return ranges and scenario weights are analytical synthesis, not price targets or individualized advice.

Facts and status labels are observed through 25 July 2026; security prices use the latest preceding market close. Recheck prices, financials, sanctions, licenses, and access rules before acting.