San Francisco Compute's CEO Evan Conrad runs GPU clusters with an anti-hoarding queue system. Small researchers get max priority by default, even over paying customers. Their Autoresearch platform uses a priority queue where you can pay to jump ahead, but you can't buy out the entire queue—solo researchers always get slots.

To kill capacity hoarding, SF Compute enforces resale requirements: if you rent compute, you must list part of it for resale. This keeps GPUs circulating instead of locked up by big labs.

Basically: prioritize the little guy, prevent monopolization, force liquidity in the GPU market.