The AI Race to Regression

Key points

• 0:00

  Most public discussions about AI capabilities are not quite on the mark.

• 0:00

  The new ChatGPT 5.0 is a significant downgrade for personal users compared to version 4.

• 0:00

  The jump from ChatGPT-4 to GPT-5 was designed for institutional users who code or design drugs.

• 0:00

  OpenAI is combining characteristics from ChatGPT-4 into ChatGPT-5 to make everybody happy.

• 1:30

  The limiting factor for artificial intelligence right now is hardware, not software.

• 1:30

  ChatGPT-4 took up about 10 terabytes of memory, while ChatGPT-5 requires at least twice or three times that.

• 1:30

  Making high-end processing chips requires 100,000 steps, 30,000 pieces, and 9,000 companies scattered around the world.

• 1:30

  The single biggest concentration of chip-making is in the United States.

• 1:30

  The number two concentration for chip-making is in the Taiwan-centric zone.

• 3:25

  If there is any degree of de-globalization, we cannot make the chips at all.

• 3:25

  Existing chips in data centers have a lifespan typically in the three to six-year range.

• 3:25

  The ability to use AI will shrink from something everyone has on their phone to simply the handful of entities capable of running their own data centers.

• 3:25

  Government efforts to reshore manufacturing only focus on fabrication facilities and ignore the entire ecosystem.

• 3:25

  We are in a moment where AI is possible now, but that will not last.

• 3:25

  We are going to see a technological regression as we lose the ability to make the hardware.

Related episodes

3:58

My Two Cents on the AI Debate

Peter Zeihan· 24 September 2026

The current debate over AI safety is actually a positive sign of open conversation. The current debate does not need to bear fruit until the 2040s when supply chains are rebuilt.

+1

3:31

The AI Race: U.S. vs. China

Peter Zeihan· 29 September 2026

China cannot compete in AI because export restrictions prevent them from building the necessary high-end hardware. The structural issues regarding data and ethics are irrelevant because the race comes down to hardware access.

4:08

The Revolution in Military Affairs: Artificial Intelligence

Peter Zeihan· 18 August 2025

AI in war will not happen as expected because the semiconductor supply chain cannot survive de-globalization. The problem will become obvious in the 2050s.

5:16

Korean Exports Skyrocket Thanks to AI

Peter Zeihan· 9 July 2026

The current AI boom relies on a fragile global supply chain involving over 50 countries. We must build everything possible now because the current system will soon fail.

4:24

The Automation of War Drags On

Peter Zeihan· 8 September 2025

Processing data is not the same as thinking. Autonomous weapon systems are simply not anywhere close.