Who Earns the Return on a Trillion Dollars of AI?

“Capex is not competitive advantage unless it becomes cash flow.”

The AI Supercycle has crossed a threshold. Global AI investment is now forecast to exceed $1 trillion in 2026. Goldman Sachs estimates roughly $580 billion of that will occur in the United States alone, while J.P. Morgan puts hyperscaler capital expenditure at approximately $697 billion this year.

These are no longer technology-sector numbers. They are macroeconomic numbers. And they force a much harder question on financial markets: who earns the return on $1 trillion of AI capital?

The market has spent three years pricing the scarcity of intelligence. It must now price the abundance of intelligence — and the scarcity of everything required to industrialize it.

1. 01 $1 Trillion: AI Has Become a Capital-Formation Event

   Goldman Sachs now forecasts more than $1 trillion of global AI investment in 2026, suggesting commonly quoted hyperscaler capex estimates alone understate the true scale of the cycle by roughly $200 billion. AI increasingly resembles the construction of a new industrial system: semiconductors → accelerated computing → data centers → power → networks → industrial AI. signal: The decisive metric is moving from AI adoption to AI capital productivity.

2. 02 $5.3 Trillion: The Financing Problem Is Only Beginning

   Goldman Sachs estimates hyperscalers could spend approximately $5.3 trillion on AI and data centers through 2030, with private markets playing an increasingly important role as the requirement expands beyond what corporate cash flows and liquid credit can comfortably absorb. AI is migrating from P&L → to balance sheet → to credit markets → to infrastructure funds → to private capital. signal: The next constraint on AI may not be technological capability — it may be the cost and availability of capital.

3. 03 35–45 GW: The Bottleneck Is Moving Into the Physical World

   U.S. data-center power demand could reach approximately 35–45 GW by 2030 — roughly twice 2024 levels. Data-center investment rose approximately 32% in 2025 and is projected by Allianz to rise another 75% in 2026. The world does not have an unlimited supply of electricity, grid connections, transformers, generation capacity, land, cooling or memory. signal: Intelligence becomes abundant; its industrialization remains intensely physical.

4. 04 3.4%: The Old Economy Still Prices the New One

   Today's U.S. CPI release is expected to show headline inflation around 3.4% year-on-year, with core inflation around 2.5%. The most capital-intensive technology transformation in modern history is occurring while the risk-free rate remains economically meaningful. Inflation → rates → cost of capital → AI infrastructure valuations → required cash flow. signal: The AI revolution may be technological, but its valuation remains financial.

5. 05 The Great Split Is Now Happening Inside AI

   The market has treated AI exposure as broadly positive. That phase is ending. The relevant distinction is increasingly between capital formers — companies generating enough cash flow to finance AI investment while maintaining strong balance sheets and acceptable returns — and capital consumers, forced to spend aggressively simply to remain competitive without yet demonstrating corresponding cash-flow creation. signal: Capex is not competitive advantage unless it becomes cash flow.

Key Takeaways

the return-on-capital argument

• AI has moved from a technology cycle into a global capital-formation cycle.
• The critical financial variable is shifting from AI spending to AI return on invested capital.
• Falling intelligence costs will not eliminate scarcity — they will relocate it toward energy, infrastructure and capital.
• The Great Split will separate companies that convert AI investment into cash flow from those forced to spend simply to remain relevant.
• The biggest winners of Industrial AI may include companies controlling the physical bottlenecks — not merely those producing the models.