AI infrastructure firm Infinity has secured $15 million at a $100 million valuation to develop a universal kernel software. This tool aims to allow AI models to run efficiently on various non-Nvidia chips, challenging the dominance of Nvidia’s CUDA platform. The company uses an automated AI agent to generate and optimize low-level code for hardware performance.
Infinity, an AI infrastructure startup, has raised $15 million in a funding round that values the company at $100 million. The investment, announced on Monday, was led by Touring Capital with participation from Principal VC and various researchers from prominent AI organizations such as OpenAI and Anthropic. The company plans to use this capital to advance its development of a universal kernel software designed to function as a competitor to Nvidia's widely used CUDA platform.
Targeting the CUDA Monopoly
Nvidia’s CUDA software has long served as a standard tool for developers, enabling GPUs to process complex AI tasks and integrate smoothly with major frameworks like PyTorch and TensorFlow. This ecosystem is a central pillar of Nvidia's market position, as it allows developers to easily deploy applications on its hardware. Infinity is attempting to disrupt this landscape by creating a universal library that abstracts low-level software complexities, allowing developers to run AI applications on diverse chip architectures including mobile processors, SRAM, and various GPU types without requiring extensive manual porting.
Automated Code Optimization
Founded by former Google Brain researcher Jeremy Nixon, Infinity utilizes a proprietary AI research agent named Ignition. This system is designed to automate the generation, testing, and debugging of low-level kernels—the core software that directs how hardware processes data. By using automation, the company claims it can optimize code for different chip architectures more efficiently than traditional manual methods. Infinity reports that its technology is already being used by AI chip developer D-Matrix, and it is currently exploring partnerships with other major cloud and hardware companies.
Business Model and Execution
Unlike traditional software firms that often rely on upfront licensing fees, Infinity has adopted a performance-based revenue model. The company charges clients based on the measured gains in performance and cost savings achieved, which are calculated by the improvement in tokens generated per second. With a current team of 26 employees, Infinity is scaling its operations as it seeks to integrate its software across the AI hardware industry. The primary investor monitorable for this startup will be its ability to scale its software compatibility across different chip architectures while maintaining code stability. Success will also depend on its ability to gain adoption among hardware manufacturers who are currently seeking to provide developers with viable alternatives to the Nvidia ecosystem.