The More You Buy, The More You Lose

Source: Where’s Your Ed At Date: 2026-07-28 URL: https://www.wheresyoured.at/the-more-you-buy-the-more-you-lose/

Summary

Ed Zitron argues hyperscaler AI capex ($1.3T+ through 2026, another trillion planned) has become self-defeating: capex now runs 24-43% of revenue and grows faster than AI revenue materializes, GPU-driven demand has pushed memory prices up 60%, and debt financing the buildout ($307B+ on-balance-sheet, $1.35T off-balance-sheet) is getting more expensive as bond spreads widen (CoreWeave issued at 540bps, now trading at 756bps). He flags OpenAI representing 69% of Microsoft’s cloud growth as evidence of underlying slowdown masked by AI spend, and calls the proposed NVIDIA-OpenAI-SoftBank circular-financing arrangement emblematic of a structure with no clear profitability path.

Implications

Feeds the AI-infrastructure-economics thread this radar tracks alongside model-capability clocks — a skeptical counterweight to the “closed labs race on throughput/product” pattern logged all week (GPT-5.6 Sol’s 750 tok/s Cerebras deal, Gemini Flash cost tiers). If the debt-cost spiral Zitron describes is real, it’s a constraint on how long capability and throughput racing can continue regardless of model quality — worth pairing with the loop’s Anthropic-IPO-financials watch item as a comparable “does the spend show up in revenue” test.