Nvidia Corp’s (NASDAQ:NVDA) Compute Unified Device Architecture (CUDA) software moat may be weakening as AI coding agents are helping to recreate similar software in just a few hours, challenging a key advantage behind its AI dominance.
Jeremy Nixon, founder of AI startup Infinity and a former researcher at Alphabet Inc.‘s (NASDAQ:GOOGL) (NASDAQ:GOOG)Google Brain, told Business Insider that the industry is reaching a turning point after his team used AI coding agents to recreate CUDA-like software for D-Matrix in 10 hours, suggesting Nvidia’s software advantage may be weakening.
Infinity is developing chip-agnostic software that helps AI models run across different hardware platforms by creating CUDA alternatives. The startup aims to simplify kernel development for AI companies while challenging Nvidia’s dominance in AI computing.
The startup is developing a universal inference library designed to work across different AI chips, enabling them to more easily reproduce cutting-edge research results.
Infinity Raises $15 Million For AI Vision
In July, Nixon told TechCrunch that Infinity was launched in 2025 to pursue "automated invention," where AI systems create and improve new technologies. The company is applying this approach to hardware by using AI to generate low-level software code that helps chips run more efficiently.
He said he created a machine learning algorithm called Omega that could generate and evaluate new algorithms through an automated feedback loop. The success led him to explore whether similar AI-driven systems could create low-level hardware code to improve chip performance.
Last month, the company raised $15 million in a funding round at a $100 million valuation, backed by Touring Capital, Principal VC, and researchers from OpenAI and Anthropic.
CUDA Still Anchors Nvidia’s AI Lead
Nvidia’s CUDA, its proprietary parallel computing platform, has been a key AI advantage for two decades. Developed by the company’s Vice President of Hyperscale and High-Performance Computing, Ian Buck, CUDA enables Nvidia GPUs to function as general-purpose processors and underpins major AI frameworks like PyTorch and TensorFlow, allowing developers to build AI applications in languages such as Python that run natively on Nvidia hardware.
In December, Nvidia’s acquisition of AI software firm SchedMD, the creator of Slurm, is seen as a move to strengthen its AI infrastructure and software ecosystem. Slurm is a key workload manager that efficiently allocates GPU resources across computing clusters, making it essential for training large AI models. Nvidia said it has become a core part of generative AI infrastructure and is optimized for the company’s latest hardware.
Futurum Group CEO Daniel Newman said that the deal could further expand Nvidia’s CUDA advantage by improving its control over large-scale AI computing workflows.
“Nvidia just deepened the CUDA moat,” he said in a post on X. “Not an easy thing to do given the moat is already eight feet deep.”
Disclaimer: This content was partially produced with the help of AI tools and was reviewed and published by Benzinga editors.
Login to comment