AI infrastructure start-up Infinity has raised $15 million in seed funding at a $100 million post-money valuation. The company is developing software that automatically prepares different AI chips to run models, reducing their dependence on Nvidia’s CUDA ecosystem.
Touring Capital participated in the round alongside Principal Venture Partners, chip-company executives and researchers from OpenAI and Anthropic. According to Infinity’s funding announcement, the capital will support product development, engineering recruitment and partnerships with semiconductor companies.
Ignition Writes Software for AI Chips
Infinity’s main product, Ignition, is an AI research agent that generates and tests the low-level software kernels required to run AI inference. Kernels control how efficiently models use a chip’s memory and computing units.
The system measures performance, identifies problems and rewrites its code to improve speed. Human engineers provide architectural direction while Ignition handles much of the testing and optimisation work.
Infinity says Ignition can adapt to GPUs, mobile processors, SRAM-based systems, systolic arrays and proprietary chip designs. The company aims to create a universal inference library that allows models to run efficiently across different hardware.
Nvidia’s position in AI computing is supported by CUDA, which connects its GPUs with software frameworks such as PyTorch and TensorFlow. Developers using those frameworks can run applications on Nvidia hardware without building separate low-level software.
Alternative chipmakers often need to create their own inference tools before customers can use their processors. Infinity is seeking to automate that work and shorten development periods from months or years to days.
D-Matrix Uses Infinity Software
AI chipmaker d-Matrix is one of Infinity’s initial customers. A published case study said Ignition reached up to 92% of the theoretical peak performance of d-Matrix’s Corsair chip within 10 hours of gaining access to the hardware.
Infinity said it had Qwen3, Qwen3.5 and Gemma4 models running on the processor within 10 days. It also reported improving inference throughput for a Qwen3-8B model beyond the performance of the widely used vLLM framework.
The company does not charge customers an upfront software licence fee. Instead, it receives a share of the cost savings and performance improvements generated by its technology, measured partly through the number of tokens processed each second.
Infinity was founded in 2025 by Jeremy Nixon, a former Google Brain researcher and co-founder of the AGI House community. The company currently employs 26 people across engineering, design and operations.
Featured image credits: Magnific.com
For more stories like it, click the +Follow button at the top of this page to follow us.