Xiaoyin Qu breaks down DeepSeek's cheap inference philosophy and the 10-month recoup calculation Liang Wenfeng runs on every model: "He explicitly said that as long as he can make his investment back, he invests in training the model, and then in 10 months, as long as he can make enough money to recoup the cost of training, which he will, then that's good for him." "Then he can put in another money to train another new model, and then make money in X months. That's his calculation." "He has…