Liang Wenfeng notes that AI coding agents are rapidly dismantling Nvidia's long-standing CUDA software moat.
Liang Wenfeng, the founder of DeepSeek, stated that the competitive advantage of Nvidia's CUDA software ecosystem is being dismantled rapidly as AI begins to take over software engineering tasks. For two decades, CUDA served as a primary barrier to entry, making it difficult for users to switch to rival hardware like AMD's. However, the rise of coding agents allows for the creation of software ecosystems much more easily than in the past. This shift is creating a significant opportunity for AMD to challenge Nvidia's dominance. While Nvidia has historically excelled in both hardware and software, AMD is known for its brute-force hardware performance. As the software barrier diminishes, the competition shifts toward hardware execution. Industry experts suggest that while AI helps rivals catch up, Nvidia continues to use AI coding agents to develop CUDA faster and validate it at a greater scale. The transition toward inference-based workloads also favors an open-source play, potentially reducing the cost of switching between different chips without rewriting software.