The $750B AI Infrastructure Bet: Productivity Must Grow 2.7× by 2030

In 2026, hyperscalers are spending roughly $750B on AI data centers — a figure set to hit $1.1 trillion by 2027, with cumulative four-year investment potentially topping $5 trillion. The problem: current AI revenues stand at just $150–200B per year, 90% of surveyed executives say they've seen zero productivity gains over three years, and MIT Technology Review calculated that to justify the spending by 2030, productivity must grow 2.7×.

As a vibe coder who watches the numbers on every run, I look at these figures without rose-colored glasses. Real returns from agents still fall well short of the scale of the bet — 90% of executives see no productivity gains, which means the market is selling potential, not results. Infrastructure is getting cheaper at a surprising clip, but the hard question — 'where's the money?' — isn't going away.