DeepSeek, the viral AI made in China and giving much-established players like OpenAI, a run for money, has a Huawei chip inside, claims a new online report. While the heavy-lifting is being done by an Nvidia, for everything that happens after, that is, thinking and responding—or inference in AI terms—the Chinese AI startup is seemingly using hardware from Huawei.
It is befitting and a stark reminder that innovation or the ability to invent and prosper cannot be curtailed. Looking at DeepSeek and Huawei, it is abundantly clear that necessity is the mother of invention. Both companies (DeepSeek obviously is new and much smaller than Huawei) have had U.S. sanctions come in the way of getting access to latest and greatest technology.
What happened with Huawei is widely known and well documented. It hasn’t been easy for DeepSeek either. In fact, things presumably have been even more challenging considering its size and scale. But necessity is what seems to have pushed the startup spun off from a hedge fund to build something—an AI—that has got the whole world talking today.
According to information shared by @Dorialexander, DeepSeek’s R1 LLM uses Huawei’s Ascend 910C chip for inference. It was trained on Nvidia H100. DeepSeek could still do better as it’s not using the fastest Huawei chip at the time of writing which could become a bottleneck for intensive workloads. But, it seems, Huawei is already working on the 920C which is billed to be at par with the Blackwell B200, Nvidia’s leading chip that is currently driving the heavyweights in the business.
