Part 8/14:
Despite the excitement, skepticism remains. Some analysts caution that the reported $6 million training cost might be misleading, as it does not factor in prior research or large-scale infrastructure investments that are typically necessary. Deep Seek’s access to extensive datasets and hardware—possibly over 50,000 Nvidia H100 GPUs—suggests they might have leveraged prior investments or hidden resources, potentially inflating the apparent cost savings.
Both possibilities hold weight: it's feasible that Deep Seek's researchers did achieve a breakthrough with cheaper hardware, but also possible that they benefited from undisclosed resources or strategic access to advanced GPUs, especially considering export control restrictions and geopolitical tensions involving China and the US.