Part I · The Streak
Chapter 02 · 13 min read

The Sunk Cost Trap and the Poisoned Well

Why intelligence is not immunity. The institutional madness of betting a balance sheet on a poisoned well.

There is a particular kind of madness that only afflicts people with access to too much capital.

The people running the great AI infrastructure empires are not fools. They are not cartoon villains pressing red buttons in glass towers, laughing while the servers burn. They are intelligent, credentialed, disciplined, and surrounded by people whose job is to make sure every decision appears rational. They have elite engineers, elite lawyers, elite consultants, elite bankers, elite lobbyists, elite communications teams, and elite risk models.

They have scenario plans, sensitivity tables, utilization forecasts, depreciation schedules, energy procurement strategies, chip supply roadmaps, and entire teams dedicated to explaining why the plan still makes sense.

The people running the great AI infrastructure empires are not fools.

In some cases, intelligence makes the trap worse. A less sophisticated person may hit a wall and stop. A sophisticated institution hits the wall, hires consultants to rename the wall, builds a framework for moving through the wall, announces a wall-optimization strategy, and raises more capital to scale the wall initiative globally.

This is how the sunk cost trap works at civilizational scale.

The sunk cost trap begins with something ordinary: commitment.

A person invests time, money, reputation, or identity into a course of action. Then evidence begins to appear that the course of action may not work. The rational response would be to evaluate the future from the present moment: ignore what cannot be recovered, assess what is true now, and choose the best next step.

[ References ]
  1. [01]
    Shumailov et al. — “AI models collapse when trained on recursively generated data, Nature 631 (2024-07-24) · www.nature.com/articles/s41586-024-07566-y
  2. [02]
    Reuters — “OpenAI and others seek new path to smarter AI as current methods hit limitations, Reuters (2024-11-11) · www.reuters.com/technology/artificial-intelligence/openai-rivals-seek-new-path-smarter-ai-current-methods-hit-limitations-2024-11-11/
  3. [03]
    Bloomberg — “OpenAI, Google and Anthropic Are Struggling to Build More Advanced AI, Bloomberg (2024-11-13) · www.bloomberg.com/news/articles/2024-11-13/openai-google-and-anthropic-are-struggling-to-build-more-advanced-ai
  4. [04]
    Microsoft — “FY25 Q2 Earnings — capex commentary, Microsoft IR (2025-01-29) · www.microsoft.com/en-us/Investor/earnings/FY-2025-Q2/