GOLDMAN SACHS PUTS A NUMBER ON THE AI BUILD-OUT: $7.6 TRILLION THROUGH 2031 — AND NOBODY CAN PROVE IT PAYS OFF Goldman Sachs has published what amounts to the most comprehensive accounting of AI infrastructure spending yet attempted, and the number it lands on is 7.6 trillion dollars. That is the firm’s baseline estimate for cumulative capital expenditure on AI compute, data centers, and power infrastructure from 2026 through 2031. In 2026 alone, the spending is projected to hit 765 billion dollars. By 2031, annual AI capex is expected to reach 1.6 trillion dollars. Breaking that down further: roughly 5.1 trillion of the total goes toward computing chips, 2.15 trillion to data centers, and 358 billion to power infrastructure. Every hyperscaler on the planet is building simultaneously, and they are doing it with debt that is now straining the US bond market. Goldman is careful to call these figures baseline estimates rather than forecasts, and it stresses that the numbers are extremely sensitive to assumptions. Change the depreciation timeline on AI chips by a few years and the total shifts by hundreds of billions. Change assumptions about compute architecture and the data center math falls apart. The firm is essentially saying: this is our best read on the trajectory given what we know, and there is enormous variance. The harder question Goldman stops just short of answering directly is whether any of this spending generates sufficient returns. A separate Goldman note from earlier this year documented what it called FOMO as the dominant investment driver, meaning fear of missing the AI wave is a stronger incentive than demonstrated return on capital. That is not an economic foundation. It is a pressure not to be left out. Seven and a half trillion dollars. The largest coordinated capital deployment in technology history. And the honest answer to whether it pays is still: we do not know yet. Keywords: Goldman Sachs AI spending forecast, AI capital expenditure 2031, AI infrastructure investment, tech capex bubble