In the past, the key resources to solve a problem were
talent + capital
Now, those resources are often
talent + capital + compute
Most of the world’s GPUs are owned by a small cohort of AI labs. Combined with more powerful AI models, this creates the potential for an extreme power differential. Overly concentrated intelligence means less space for idealogical diversity.
Most companies (especially high-variance research and early-stage companies) that are building or deploying AI/ML models are blocked by access to GPUs.
We can encourage a more democratic intelligence landscape by distributing computation among more players.
This is a tractable economic problem. We’re starting with a small contribution, designed to scale up a lot.