GPU World

The post argues that the AI revolution has barely begun: billions of people have never interacted with a frontier LLM because compute limitations mean only a few million GPUs capable of serving such models are manufactured each year, leaving humanity 'GPU-poor'. It then proposes a thought experiment: by 2040, enough GPUs could exist to provide every person with the performance equivalent of a B300 GPU for contemporary LLMs.

The premise: frontier AI progress stops as of September 2026 — models get faster and cheaper but never become superhuman or broadly improve. The Singularity never happens, but GPU manufacturing continues, leading to roughly 8 billion GPUs globally. The contest asks readers to explore the consequences: constant AI surveillance, infinitely patient tutors, the end of social media as we know it, world-class AI doctors and personalized medicine, and impacts on the developing world.