Astral Codex Ten

GPT-7

3 min read

GPT-7

Article

GPT-7 is a recurring concept in the Astral Codex Ten archive, appearing 2 times across 2 issues between February 13, 2024 and November 26, 2025. The archive places it in contexts such as “GPT-7 to cost $2 trillion”; “GPT-7 might need all of the world’s computers”; “you can spend ten years training GPT-7 very slowly”. It most often appears alongside Google, GPT-6, Microsoft.

Metadata

  • Category: Concepts
  • Mention count: 2
  • Issue count: 2
  • First seen: February 13, 2024
  • Last seen: November 26, 2025

Appears In

Source Context

Recovered passages from the original issue text. When the raw archive preserved outbound links inside the source passage, they are listed directly under the quote.

Sam Altman Wants $7 Trillion
February 13, 2024 · Original source
So each GPT costs between 25x and 100x the last one. Let’s say 30x on average. That means we can expect GPT-6 to cost $75 billion, and GPT-7 to cost $2 trillion.
If we keep our 30x scaling factor, GPT-5 will take 1/70th of all the computers in the world, GPT-6 will take 1/2, and GPT-7 will take 15x as many computers as exist. The computing capacity of the world grows quickly - this source says it doubles every 1.5 years, which means it grows by an order of magnitude every five years, which means these numbers are probably overestimates. If we imagine five years between GPTs, then GPT-6 will actually only need 1/10th of the world’s computers, and GPT-7 will only need 1/3. Still, 1/3 of the world’s computers is a lot.
Inline links: this source
GPT-4 took about 50 gigawatt-hours of energy to train. Using our scaling factor of 30x, we expect GPT-5 to need 1,500, GPT-6 to need 45,000, and GPT-7 to need 1.3 million.
Inline links: 50 gigawatt-hours
Why AI Safety Won't Make America Lose The Race With China
November 26, 2025 · Original source
China is relying on this. They know they can’t compete on the compute and model layers in the near-term6, so they’re hoping to win on applications. They imagine America having a slightly better model - GPT-7 instead of GPT-6 - but our GPT-7 is sitting in a data center answering user questions and generating porn, while their GPT-6 is helping to run schools, optimize factories, and pilot drones. America’s task isn’t micro-optimizing our already large compute and model advantages - gunning to bring the score to GPT-7.01 vs. GPT-6. It’s responding to the application-layer challenge that China has set us.
Inline links: 6