Arthur Hayes: FLOP miner block rewards are distributed based on the proportion of inferred computing power
Arthur Hayes, co-founder of BitMEX, tweeted that many retail and corporate FLOP miners are asking about the best hardware choices and ways to optimize hardware configurations, but Flop Labs cannot answer these questions. Block rewards are distributed based on the amount of inference processed, weighted by floating-point operations; if a miner processes 100 GFLOPs out of a total of 200 GFLOPs in a block, they will receive half of the miner reward for that block.
Arthur Hayes stated that if the purchased hardware can only run under-demand AI models, it will not be able to handle any inference requests and will not generate revenue. He suggests referring to the most popular models on OpenRouter, predicting the types of models that will be in highest demand in agent scenarios, or first testing the existing configuration and then adjusting based on actual results.






