Hyperscalers have sunk over $800 billion into AI infrastructure over three years, with another $700 billion planned for 2026 and $1 trillion in 2027. Nobody, including Microsoft, Google, Amazon, or Meta, can publicly explain how these investments break even. Every AI startup bleeds cash, and the math for the industry as a whole has not added up on any realistic timeline.

Microsoft alone has spent roughly $100 billion on its OpenAI partnership, per testimony during the Musk-OpenAI trial. Since FY2023, Microsoft's capex hit $293.8 billion, with about 30% directed at OpenAI infrastructure. In FY2025, Microsoft booked an estimated $17.9 billion in AI revenue—less than a fifth of what it spent building capacity.

Even generous revenue projections don't close the gap. Anthropic's rumored $45 billion annualized revenue would still fall short of recovering a single year of hyperscaler capex. Microsoft 365 Copilot generates a maximum $7.2 billion at full price. Without AI revenues exploding, capex halting, and margins turning positive simultaneously, the best outcome is a few hundred billion dollars of waste.