What Should GPUs Really Do? | MOONSHOTS

The transcript presents a compelling economic argument about the future allocation of GPU computing resources. The speaker raises concerns about the viability of self-driving cars in a world where GPU compute becomes increasingly valuable for other applications. The core issue identified is that autonomous vehicles require substantial computational power - essentially a full GPU per vehicle - but this same computational capacity could be deployed for potentially more valuable purposes. By the end of the year mentioned, the speaker suggests that GPU technology will advance to enable breakthrough applications in medicine (specifically brain surgery) and scientific research (discovering new mathematics and physics). This creates a resource allocation dilemma where the relatively mundane task of transportation may not be able to compete economically with these high-value applications. The fundamental tension described is between using finite GPU resources for everyday convenience versus transformative scientific and medical breakthroughs. As demand for compute power approaches infinity, the speaker suggests that driving people around may not meet the economic threshold to justify the computational cost.