FT: Only about a third of US $10tn AI infrastructure projects may be realistic
lemire · x · 2026-10-19
The Financial Times reports that America's roughly $10 trillion AI infrastructure boom is starting to fray, with maybe only a third of large AI infrastructure projects realistically feasible.
Key bottlenecks include:
- Grid connection delays
- Transformer shortages
- Liquid-cooling supply constraints
- Local community opposition
Commenting on the report, lemire notes that if this holds, space data centres could become far more attractive in the coming years.
Related event: US AI Infrastructure Boom Cools as 47 Major Data Center Projects Canceled(3 posts)→
More from Infra
- Sort prompts before batch MoE inference: a free 7–17% speedup in stock vLLM — yb2698 · 2026-10-20
- GPU warehouses to be dumped in '27-'28? Blogger bets 90% AIDCs still get built — teortaxesTex · 2026-10-20
- Compute bubble spar: datacenters will get built, but expect 'funny maneuvers' in the supply chain — jobergum · 2026-10-20
- BofA: AI capex running 50% above forecasts, hyperscaler cost of capital jumps 4% to 10% — rohanpaul_ai · 2026-10-20
- Morgan Stanley: 2027 AI capex hits $1.58T with leases counted; Nvidia takes 44% — rohanpaul_ai · 2026-10-20
- PFN and Tenstorrent win NEDO project to develop optically integrated xPU through 2031 — DavidBennett__ · 2026-10-20