AI’s Next Bottleneck Can’t Be Bought

What You’ll Learn From This Edition

• Why the shift from chip scarcity to power scarcity is a change in kind, not degree, and why that distinction matters more than any gigawatt figure

• A reusable test for telling whether a resource in any value chain is market-cleared or permission-cleared, and why that test predicts who wins

• Which unglamorous companies are quietly becoming as strategically important to AI as the labs themselves

• Why ordinary electricity ratepayers are now a stakeholder in the AI race in a way chip buyers never were

• How to read the next eighteen months of AI infrastructure announcements without being fooled by megawatt numbers

• A framework for where founders, consultants, and investors should actually place bets as this shift plays out

Table of Contents

1. Executive Summary

2. Deep Dive in One Sentence

3. Why This Topic Matters Now

4. The Big Question

5. The Conventional Narrative

6. What’s Really Happening

7. The Economics Behind the Shift

8. Winners and Losers

9. Second-Order Effects

10. Strategic Implications

11. Mental Model of the Week

12. Key Takeaways

13. Closing Thought

Executive Summary

• Chip scarcity was a market problem, solvable with capital: pay more, order earlier, fund more fabs. Power scarcity, at the point an AI data center actually needs it, is a permission problem, solvable only through queues, hearings, and public consent.

• US grid interconnection queues hold roughly 2,000 to 2,600 gigawatts of stuck projects with average wait times near five years, and even the turbine industry’s own executives say equipment is not what’s gating buildouts.

• Value is fragmenting away from a single chokepoint supplier toward old-economy incumbents: turbine makers, merchant generators, nuclear operators, and energy-rich jurisdictions.

Read more