The Information Machine

Is AI Demand a Structural Shift or a Hype Cycle?

Synthesis history

13 versions, newest first.

  1. Version 13 2026-06-03 18:49 UTC · 177 items

    Three additions extend the quantitative scale of the bull case. Morgan Stanley data shows Google, Amazon, Microsoft, and Meta on track to spend $1 trillion on AI infrastructure in a single year by 2027 — up from $250 bi…

  2. Version 12 2026-06-02 08:26 UTC · 168 items

    Three additions extend the bull camp's reach this pass. Jensen Huang escalated from bullish on Nvidia's market cap to bullish on a competitor's revenue, calling Amodei's $1T forecast for 2030 'too conservative' and pred…

  3. Version 11 2026-06-01 02:36 UTC · 161 items

    Three developments add new texture this pass. Multiple mainstream sources—Forbes, Fortune, AI Weekly—now explicitly name the Solow Paradox as the historical frame for AI's failure to show economy-wide productivity gains…

  4. Version 10 2026-05-30 18:34 UTC · 149 items

    Three meaningful new developments this pass. SemiAnalysis's 'Dark Output' thesis introduces a structural measurement explanation for why 80%+ of surveyed executives report no productivity gains despite genuine AI spendi…

  5. Version 9 2026-05-27 02:42 UTC · 144 items

    Two meaningful new developments this pass. First, Meta has adopted Amazon's custom AI chips [^21150]—a qualitatively more significant signal than the earlier Uber-Trainium3 report because Meta is itself a hyperscaler an…

  6. Version 8 2026-05-25 20:14 UTC · 132 items

    Two developments deepen existing threads without introducing wholly new fault lines. First, analysis now projects CoWoS, HBM, and 2-3nm capacity constraints extending through 2027 [^20634], lengthening the semiconductor…

  7. Version 7 2026-05-25 12:24 UTC · 123 items

    Two new developments deepen existing threads without introducing wholly new fault lines. First, reporting on the House subcommittee hearing [^19897] reveals specific policy outcomes: bipartisan agreement on data center …

  8. Version 6 2026-05-25 06:29 UTC · 116 items

    Two substantive new developments this pass. First, Uber's reported selection of AWS Trainium3 over Nvidia hardware at approximately 50% lower cost [^18741][^18742] provides the first concrete enterprise-level signal for…

  9. Version 5 2026-05-24 19:14 UTC · 112 items

    New items this pass deepened two existing themes without introducing new fault lines. On the power grid side, the RMI's framing of the interconnection queue as a structural barrier with its own reform dynamics [^16068],…

  10. Version 4 2026-05-24 10:54 UTC · 100 items

    Two significant additions entered this pass. First, the data center delay story has been quantified: multiple reports now indicate 30–50% of planned 2026 US data centers are facing delays or cancellations due to power g…

  11. Version 3 2026-05-23 04:46 UTC · 70 items

    Three material developments entered the record this pass. First, the thread's most active open question—whether the mid-2025 CoWoS utilization gap would close—has been answered bullishly: Nvidia has secured roughly 60% …

  12. Version 2 2026-05-22 18:46 UTC · 53 items

    Three material developments entered the record this pass. First, TSMC CoWoS utilization data (reportedly 60% as of August 2025 [^10467]) provides the first concrete empirical test of Baker's TSMC-as-bubble-signal thesis…

  13. Version 1 2026-05-22 08:13 UTC · 4 items

    The debate over whether AI demand is a structural economic shift or a hype cycle is being contested with first-party data and investor frameworks. SemiAnalysis has published internal token-spend data showing 10–90x ROI …