🤖 AI Summary
Economists warn that American growth in H1 2025 is shockingly narrow: Jason Furman’s analysis finds 92% of GDP growth came from investment in data centres and information-processing equipment — a sector that is only about 4% of the economy. Strip out those infrastructure spends and headline growth collapses to roughly 0.1% annualized. That concentration creates an economic “monoculture”: labour markets, retail, real estate and manufacturing contributed almost nothing, so any slowdown in AI infrastructure buildout (from regulatory, technical or financing shocks) could quickly tip broader activity toward recession.
The technical and financial mechanics deepen the worry. Big chip and cloud players are entangled in circular financing — e.g., Nvidia’s reported up-to-$100B exposure to OpenAI, a 7% stake in CoreWeave (~$3B) while CoreWeave has bought ~$7.5B in Nvidia GPUs, and multi‑billion dollar capacity and rental commitments — that inflate revenues and lower borrowing costs (startups’ rates falling from ~15% to 6–9%). Meanwhile hardware depreciates rapidly (Harris Kupperman estimates ~$40B annual depreciation vs $15–20B revenue for new centres), capex is projected to jump from $375B to $500B in 2026, and power needs could rise 165% by 2030, with ~$720B in grid upgrades required. History (Railway Mania, late‑90s telecom) shows infrastructure can outlast bubbles, but rapid obsolescence, circular finance and strained grids make the current AI buildout a high-reward yet high‑fragility gamble.
Loading comments...
login to comment
loading comments...
no comments yet