AI Data Centers Are an Even Bigger Disaster Than Previously Thought (futurism.com)

🤖 AI Summary
Harris “Kuppy” Kupperman, founder of Praetorian Capital, published a follow-up to an essay arguing that the economics of AI data centers are far worse than commonly portrayed. After conversations with two dozen senior data-center professionals he concluded his original 10-year depreciation assumption was wildly optimistic: physical infrastructure and especially semiconductor accelerators are being cycled or rendered obsolete within roughly 3–10 years due to rapid technology turnover and constant, high-power use. Using revised assumptions, he now estimates the industry would need $320–$480 billion in 2025 revenue (versus an earlier $160 billion estimate) just to break even on that year’s capex, and roughly $1 trillion when accounting for 2025–26 buildouts — while current AI revenue is nearer $20 billion annually. The significance for AI/ML is stark: the sector is far more capital-intensive and short-lived than investors or operators realize, meaning hardware churn, replacement costs, and tight gross-margin assumptions (Kupperman used a 25% gross margin in his math) could make profitability unattainable at scale. He warns that scaling doesn’t fix the math — it amplifies financial strain — turning an industry problem into a potential macroeconomic risk if dozens or hundreds of new data centers come online without matching revenue. The findings prompt urgent questions about sustainable deployment models, hardware lifecycles, and where future AI value and returns will actually come from.
Loading comments...
loading comments...