Infinity, an AI infrastructure startup, recently secured $15 million in funding at a $100 million valuation. The round attracted investment from notable backers including Touring Capital, Principal VC, and researchers affiliated with OpenAI and Anthropic. Founded last year by former Google Brain researcher Jeremy Nixon, Infinity aims to create software that simplifies running AI models across various AI chips, challenging the dominance of Nvidia’s CUDA software, which currently underpins most AI frameworks like PyTorch and TensorFlow.

The company is developing a universal inference library designed to operate on a wide range of hardware architectures, including SRAM, GPUs, phone chips, and systolic arrays. Infinity’s key innovation is its AI research agent called Ignition, which autonomously generates and refines low-level kernel code needed to optimize AI model performance on different chips. This self-optimizing system tests, debugs, and rewrites code to enhance hardware efficiency, potentially replicating Nvidia’s software stack functionality but for a wider array of processors.

Infinity's business model involves partnering with chip manufacturers rather than charging upfront licensing fees. Instead, it earns revenue by taking a share of the improvements in performance and cost savings it helps generate. Among its current customers is D-Matrix, an AI chip company positioning itself as an Nvidia alternative. Infinity is also negotiating with other major chip and cloud providers, signaling ambition to expand its footprint in an ecosystem dominated by Nvidia hardware.

Jeremy Nixon’s vision for Infinity stems from his previous work on an autonomous machine learning algorithm he created called Omega, which could invent and evaluate new algorithms automatically. Carrying this “automated invention” concept into hardware software, Infinity is pushing to automate the creation of efficient code for AI inference chips. With a 26-person team, the startup is part of a growing movement seeking to promote hardware and software alternatives that reduce reliance on Nvidia and enable broader AI hardware competition.