Nvidia’s competitive edge is facing new challenges as artificial intelligence begins to automate the creation of software like its CUDA architecture. CUDA, developed over years by Nvidia executive Ian Buck, includes ready-to-use code for AI tasks, debugging tools, and chip synchronization software. Jeremy Nixon, founder of AI startup Infinity and former Google Brain researcher, told Business Insider his startup used AI coding agents to replicate CUDA-like software for chip startup D-Matrix in just 10 hours, indicating Nvidia’s barrier is eroding. Major cloud services including Google, Amazon, and Microsoft have invested in software for their AI chips, while OpenAI and Anthropic have shown AI models generating system software. DeepSeek founder Liang Wenfeng said coding agents and its proprietary language TileLang streamlined AI software development. Nvidia reports rising developer use of CUDA libraries, and Vice President Ankit Patel said the company uses AI coding agents for faster CUDA development. CUDA’s secondary advantage is its extensive application ecosystem, creating a lock-in effect, with Amazon internal communications identifying CUDA as an impediment to adopting its Trainium and Inferentia chips. Chris Lattner, CEO of Qualcomm-owned Modular, said CUDA’s legacy framework, originating in gaming, is both an asset and a limitation.