DeepSeek has claimed incredible profitability for its AI models. But is it really so?

Chinese artificial intelligence startup DeepSeek has shocked the market by announcing a theoretical profitability of 545% for its V3 and R1 models. According to the company, such figures are possible if only the current costs of query processing (inference) are taken into account and ignore the costs of training neural networks and scientific research. However, experts doubt it: the real margin is likely to be many times lower. Why is 545% just a theory? DeepSeek explained that the calculations are based on one day’s data at the end of February, when demand for their services was high. But in reality, the company faces a number of limitations. For example, most of its services are not yet revenue-generating, and during off-peak hours […]

Material DeepSeek has claimed its AI models are incredibly profitable. But is it really true?” was first published on ITZine.ru.