AI Demand Begins to Justify Massive Cost of Data-Center Buildout

What happened

A Bloomberg analysis shows global AI revenue (excluding China) reached roughly $25B in Q1 2026, exceeding for the first time the estimated $21B in depreciation costs from AI data-center buildout.

Context and impact

This is an important data point in the ongoing 'AI ROI is a myth' debate. So far the most aggressive numbers have come from Sequoia ($600B revenue gap), Goldman Sachs, and David Cahn — all pointing to massive compute spending without matching revenue. Bloomberg shows the crossover has happened — but barely. Any demand shock (recession, regulation, cheap open-weight substitutes) could break the balance. It lands alongside the 'chip-wreck' sell-off and Wall Street's growing AI backlash worry — the narrative is fragile.

Details

• AI revenue Q1 2026 (ex-China): ~$25B
• AI buildout data-center depreciation: ~$21B
• Crossover is narrow, margins thin
• Lands alongside 'chip-wreck' and Bloomberg's 'Maybe the AI honeymoon is over'
• Risk: open-weight models (GLM-5.2, DeepSeek) pushing inference prices down