1. A New World. An Old Economic Law.

Artificial intelligence can change the world. A bond yield can change what that future is worth.

Governments are rebuilding strategic industries. Companies are committing enormous sums to artificial intelligence, electricity networks, semiconductor manufacturing and data centers. Much of that infrastructure will consume capital years before it earns an adequate return. The investment case cannot stop at technological potential.

In our previous Investment Intelligence analysis, The Return of the Cost of Capital, we examined the broader change in financing conditions. Now we follow the mechanism into an investment decision, then test capital productivity against two audited corporate records.

Growth alone is not enough. An extraordinary business can destroy value through poor reinvestment; indispensable infrastructure can earn inadequate returns. A technological revolution can succeed while many of its investors lose money. The mathematics have not changed. The environment in which they operate has.

2. The Federal Reserve Does Not Control the Entire Price of Money

Investors obsess over the Federal Reserve: twenty-five basis points higher, twenty-five lower. Policy matters, but a company’s long-term cost of capital is not simply the Fed’s policy rate. Treasury yields incorporate expectations about real rates, inflation, future policy and the compensation investors demand for holding duration. The dollar transmits American financial conditions internationally without making the ECB, Japan or China’s financing systems identical.

Geopolitics can alter that pricing. An energy disruption around the Strait of Hormuz could affect transport costs, inflation expectations and required returns. Sovereign financing needs can also influence market clearing yields. These relationships are conditional: weaker growth can pull yields down even as uncertainty rises.

On December 30, 2020, the US 10-year Treasury yield was 0.93%. On October 14, 2026, it was 5.19%. These are nominal dollar benchmarks, not inflation-adjusted returns or company borrowing quotes. Credit risk, maturity, currency and liquidity shape borrowing costs; equity investors require compensation for ownership risk. The weighted average cost of capital, or WACC, combines the required equity return and after-tax debt cost using an appropriate financing structure.

Take a hypothetical company with 70% equity, 30% debt, a 5.19% risk-free reference, a 5% equity risk premium, beta of 1.10, a 6.2% assumed debt cost and a 25% tax rate. Its modeled equity cost is 10.69% and WACC is 8.878%, rounded to 8.9%. The debt cost is a scenario assumption, not a reproduced corporate-bond index observation. None of these inputs estimates Amazon or Visa’s WACC.

3. The $100 Million Decision

Imagine investing $100 million today for $16 million in annual free cash flow, paid at each year-end for ten years, with no residual value. At an 8.9% discount rate, net present value is positive $3.14 million. At 10.9%, it is negative $5.38 million. An $8.51 million swing, without changing a single projected cash flow.

The factory still operates, the technology still works and customers still pay. But the project no longer clears its economic hurdle. This is a sensitivity exercise, not evidence that Treasury yields mechanically pushed WACC up by two percentage points. Cash-flow expectations, risk premiums and financing structures can change together. Higher yields do not inevitably destroy value; they force the investor to reconsider what justifies the commitment.

4. AI Is Becoming an Industrial Economy

AI depends on an industrial system: electricity, grid connections, cooling, semiconductors, data centers, networks, models and applications. Each layer has its own capital requirements, competitive pressures and risks. The International Energy Agency estimates data-center electricity consumption at 485 TWh in 2025 and projects 950 TWh in 2030 in its updated central outlook. That is a projection for all data centers, not guaranteed AI demand.

The financing questions are immediate. Who pays for construction? Who bears utilization risk? How quickly does equipment become obsolete? Which contracts protect cash flows? A successful AI industry does not guarantee attractive economics for every participant financing its expansion.

5. Geopolitics Is Changing the Economics of Efficiency

Global production networks created enormous value through specialization, scale and efficiency. They also created dependencies. Semiconductors, critical minerals, energy transport and digital infrastructure now carry strategic importance beyond corporate profitability. US–China competition can encourage selective fragmentation, additional domestic capacity, larger inventories and duplicated infrastructure.

Those commitments may be rational for governments while offering unattractive returns to private investors. Resilience has economic value, but someone must pay for it. Strategic necessity does not eliminate the investor’s obligation to examine cash flows and the price paid.

6. The Next Scarcity May Not Be Intelligence

If AI becomes cheaper and more widely accessible, competition may transfer substantial value to customers. Profits could instead accrue to scarce complements: reliable electricity, grid access, engineering expertise or specialized equipment. Established businesses may also benefit by using inexpensive intelligence to improve capital productivity.

An industrial company that reduces downtime, releases working capital and expands output without proportionately expanding its asset base can improve returns without extraordinary revenue growth. That is a different proposition from financing another capital-intensive facility. The United States, China, East Asia, Europe, India, the Gulf and the Nordics offer different combinations of energy, capital, technology and institutional strengths. Geography alone establishes no investment merit. As autonomous systems enter operational infrastructure, cybersecurity, authorization and control may also become valuable capabilities.

7. Growth Is Not Value Creation: Amazon and Visa

Amazon combines commerce, logistics, advertising and cloud infrastructure. Visa runs a global payments network. Between fiscal 2020 and 2025, their revenue indices rose from 100 to 185.7 and 183.1 respectively. Almost the same growth; very different capital demands. By 2025, Amazon’s gross cash property and equipment purchases represented 18.4% of revenue, against 3.7% for Visa’s property, equipment and technology purchases. These measures exclude acquisitions and noncash lease additions. Amazon closes its year in December; Visa in September.

In 2025, Amazon generated $139.5 billion of operating cash flow and spent $131.8 billion on gross cash capital expenditure, leaving $7.7 billion on a simple operating-cash-flow-minus-capex definition. Visa generated $23.1 billion and spent $1.5 billion, leaving $21.6 billion. This comparison uses gross purchases consistently; Amazon’s published free cash flow instead nets asset sales and incentives and was $11.2 billion. Neither definition is unlevered free cash flow to the firm.

Capital expenditure still cannot settle the question. Our tax-normalized ROIC divides reported operating profit after a common assumed 25% tax charge by average beginning- and end-year invested capital. We retain goodwill and acquired intangibles, treat operating leases as operating expenses, include finance-lease debt and exclude cash, quoted securities and identified Amazon financial investments. On that basis, 2025 ROIC was approximately 22.1% for Amazon and 42.2% for Visa.

Visa demonstrates stronger measured capital productivity in 2025 under this definition. The result survives a more conservative Amazon denominator retaining its financial investments: Amazon’s return falls to 19.3%. Removing goodwill raises both companies’ measured returns and preserves Visa’s lead. That ranking is not true in every year: the two were close in 2020. Expensed development, acquisition accounting, cash requirements, litigation and depreciation estimates prevent a perfectly neutral comparison.

Amazon’s 2020–2025 increase in tax-normalized operating profit divided by the increase in identified invested capital is about 17.9%. This is a backward-looking incremental-return proxy, influenced by the recovery of existing operations, not the return earned by a new AI facility. We omit Visa’s corresponding percentage: its small net capital increase makes that ratio unstable and economically misleading.

For context, assumption-based October 2026 WACC ranges are 10.5–12.5% for Amazon and 9.0–10.5% for Visa. These are sensitivity estimates using different beta ranges, not measured market WACCs or historical hurdle rates. Compared with the defined 2025 ROIC measures, these illustrative WACC ranges imply positive ROIC–WACC spreads for both companies. This is a conditional sensitivity comparison, not evidence of realized economic profit. A higher return percentage also does not establish a larger dollar contribution or a better investment at today’s stock price.

The practical distinction is reinvestment burden. Similar top-line growth can leave radically different amounts of cash available after physical investment. Visa’s economics offer more room for distributions; Amazon commits more cash to capacity whose future returns still have to be earned. Amazon’s consolidated figures cannot isolate AI returns. Nor does Visa’s low physical capital intensity make its shares inherently superior. An exceptional company can be an unattractive stock at an excessive price.

8. The Next Twenty Years Will Reward Judgment

Artificial intelligence, energy infrastructure, advanced manufacturing and autonomous systems may reshape the world. Economic progress does not automatically produce attractive returns for those financing it. An investment earning 15% on capital against a 9% required return can create value; one earning 7% against an 11% hurdle can grow while destroying it. These are illustrative businesses, not Amazon or Visa estimates.

The determining questions remain concrete. What cash flows will the investment generate? How much capital will it consume? How durable are its advantages? What risks does the investor bear? How much expected success is already reflected in the price? The bond market cannot tell us which technology will dominate, but it influences the financial terms on which that future gets built.

The next twenty years will create extraordinary winners. They will also expose the difference between owning a compelling story and owning a productive economic asset. The future will require enormous capital. The winners will be those who know what that capital is worth.