DeepSeek's move to develop its own inference chip has drawn market attention not so much for short-term order changes, but for the evolving long-term pricing logic of the AI computing power industry chain.
In the past, AI competition primarily revolved around model training, with high-performance GPUs holding a core position thanks to the CUDA ecosystem. However, as AI applications enter the large-scale inference stage, the importance of cost, energy consumption, and deployment efficiency has increased, giving ASIC chips optimized for specific models a competitive edge. Companies like OpenAI, Google, and Amazon are developing custom chips, essentially aiming to reduce their reliance on general-purpose GPUs.
For Nvidia, current training demands and its high-end computing power advantage remain solid, and DeepSeek's self-developed chip cannot directly replace its core market. However, whenever a cutting-edge lab (such as OpenAI, Anthropic, and DeepSeek) or a large cloud computing company (such as Google and Amazon) decides to develop its own dedicated chip, it weakens the argument that "Nvidia will always be the only choice in the field of artificial intelligence."
