AI infrastructure company Infinity has announced the completion of a $15 million funding round, bringing its post-money valuation to $100 million. Investors include Touring Capital, Principal VC, and researchers from OpenAI and Anthropic. The company is developing a software layer for AI inference aimed at making it easier for different types of chips to run models.
Product direction corresponding to financing
In today’s AI supply chain, NVIDIA’s advantage stems not only from chip performance but also from its CUDA software ecosystem. Developers typically build applications using mainstream frameworks like PyTorch and TensorFlow, which have long been built on CUDA, causing the majority of AI applications to run by default on NVIDIA chips.
Infinity aims to address exactly this dependency. The company is trying to build a lower-level software alternative to CUDA, making it easier for a wider range of chips to handle AI inference tasks. According to the company, this software can be adapted to various architectures, including GPUs, mobile chips, SRAM, and systolic arrays.
Targeting cross-chip inference capabilities
Infinity aims to build a universal inference library that enables different chips to run and reproduce cutting-edge AI research. For most application-layer startups, writing low-level kernels and migrating applications to other chips is typically time-consuming and lacks sufficient engineering resources—this is one of the key reasons why Nvidia’s software ecosystem has long held an advantage.
Company founder Jeremy Nixon, formerly a researcher at Google Brain and the creator of the AGI House community, said the idea for Infinity stemmed from "automated invention." He believes AI systems can not only generate models but also create the underlying code that enables chips to run more efficiently.
Generate underlying code using AI agents
One of Infinity's core products is an AI research agent called Ignition. This system writes the low-level code required for inference on non-Nvidia chips, and handles testing, debugging, and performance measurement. If the results are unsatisfactory, the system automatically rewrites the code to further improve runtime efficiency.
According to the company, this system has continuous self-optimization capabilities and is compatible with various chip architectures, including hardware with proprietary designs. Infinity positions itself as a software stack approaching the CUDA level.
Currently, Infinity’s customers include AI chip company D-Matrix. Jeremy Nixon said the company is also in discussions with other major chip manufacturers and cloud computing firms. Its business model does not involve upfront software licensing fees; instead, it shares in performance improvements and cost savings, measured by metrics such as tokens processed per second.
Additional information: As of now, the Infinity team consists of 26 members covering design, operations, and engineering roles.