Ben Bajarin: $9.6T AI capex vs $4.1T revenue — AI pays for itself, but later than you think
BenBajarin · x · 2026-11-07
Ben Bajarin analyzes whether AI infrastructure can fund itself. The bear case: $9.6T of AI datacenter spending in 2024-2030 vs $4.1T of trackable AI revenue in 2026-2030. But he argues the comparison is wrong — hardware is expensed over its useful life while AI revenue is recurring and compounding. The fairer test is whether each year's revenue covers depreciation, power and capital return; on that basis AI does pay for itself, just later than optimists hope. A rented gigawatt repays build cost in 3 years; the harder problem is absorption — excess capacity trains models or powers bundles that never show up as AI revenue.
More from Venture
- Investor cites FT: public investors missing the AI narrative are "failing a basic IQ test" — firstadopter · 2026-11-07
- AI startups should ship now: frontier labs are quiet, attention is up for grabs — kimmonismus · 2026-11-07
- OpenAI secures $11 billion in new funding at a $1.25 trillion valuation — Polymarket · 2026-11-07
- Enterprise AI Spending Up Nearly 5x in a Year, Driven by Agents — The AI Daily Brief · 2026-11-07
- Theo earned $14,791 in one X payout cycle: 220 posts, 35M views dissected — MatthewBerman · 2026-11-07
- OpenAI's round shrank from $30B ask to $11B raised — Anthropic IPO looms — kevinsxu · 2026-11-07