Columbia report: 182.7 GW of US data centers must yield $3.7T annual revenue by 2032
rohanpaul_ai · x · 2026-09-27
A Columbia Business School report, "Financing the AI Buildout," puts numbers on the $800B–$1T AI capex wave:
- Its central scenario is 182.7 GW of additional US data-center capacity between 2025 and 2032
- Using GB300 NVL72 configurations, that equals 1.07 million racks / 77 million GPUs
- The buildout requires $3.725 trillion in mature annual revenue — roughly $5.5 per installed GPU-hour at full utilization
- That sits below current $6–$10+/GPU-hour pricing for high-end NVIDIA capacity, making the central scenario economically plausible
The report frames the wave as hyperscalers contributing credit quality, outside investors providing capital, and frontier AI firms gaining compute they couldn't finance alone.
More from Venture
- XPRIZE's $2.5M Future Vision winner becomes first prize project the public can invest in — PeterDiamandis · 2026-09-27
- Big 4 AI capex hits 2.4% of GDP, double the telecom bubble peak — but still below railroads — rickasaurus · 2026-09-27
- Goldman: token demand to grow 18x by Sept 2026, but frontier demand lags at 8-9x — rohanpaul_ai · 2026-09-27
- Banks price orbital compute at $165-180B per GW; new model says $68B by 2028 — JOBhakdi · 2026-09-27
- Personal AI assistants now cost $3k-$7k per user per year — here's where startups should attack — illscience · 2026-09-27
- GLM-5.3 Flash Matches Claude at 1/429th the Price in a YouTube Script Benchmark — OnlyProggingForFun · 2026-09-27