AI Is Too Expensive: Hyperscalers Need $3 Trillion to Break Even

Ed Zitron argues that AI, in its current state, is not economically viable for anyone except hardware vendors like NVIDIA. Hyperscalers (Microsoft, Meta, Amazon, Google) have collectively invested over $800 billion in AI capex over the last three years, with plans to add another $700 billion in 2026 and $1 trillion in 2027. This means they need to generate at least $3 trillion in AI-specific revenue just to break even — and $6 trillion for a worthwhile return.

Microsoft alone has spent approximately $100 billion on its OpenAI partnership (including investments, infrastructure, and hosting costs), per testimony from an executive during the Musk-OpenAI trial. This represents about 30% of Microsoft's total capex since FY2023 ($293.8 billion). Microsoft's entire AI revenue for FY2025 is estimated at ~$17.9 billion — less than a fifth of its capex. Even its best reported figures (e.g., $37 billion AI revenue run rate) are snapshots from a single month, not annual projections.

The article notes that Microsoft's 20 million Microsoft 365 Copilot subscribers yield at most $7.2 billion in revenue (assuming $30/month per user, but discounts are common). A more realistic figure from FY2025 is ~$7.5 billion from OpenAI inference spend plus $761 million in revenue share.

The bottom line: current AI revenue streams are dwarfed by the infrastructure spend. Until costs drop or usage economics shift dramatically, the AI buildout is a financial gamble, not a sure bet.