Deepseek story was way overblown. Read the gpt-oss paper, the actual training run is not the only expense. You have multiple experimental training runs as well

Deepseek story was way overblown. Read the gpt-oss paper, the actual training run is not the only expense. You have multiple experimental training runs as well as failed training runs. + they were behind SOTA even then