Infinity raises $15M to make AI chips inference-ready

AI chip startups do not only compete on silicon. They also need the software that lets real models run efficiently on their hardware, and that layer can take months of specialist engineering.

What happened

Infinity raised a $15M seed round to develop Ignition, an autonomous software agent that writes and optimises the low-level inference software required to run AI models on different processors.

The company is targeting work normally handled by highly specialised compiler, kernel and performance-engineering teams. These engineers adapt models to a chip’s architecture, memory layout and execution patterns so the hardware can deliver usable speed and efficiency.

Infinity is already working with AI-chip company d-Matrix. Its wider ambition is to make new processors usable much faster by automating much of the optimisation work between finished silicon and a production-ready AI workload.

Why it matters

A technically capable chip can still struggle commercially if developers cannot deploy models on it easily. Nvidia’s advantage is not only its processors; it also has a mature software ecosystem that reduces friction for customers.

That makes software readiness one of the biggest barriers facing alternative AI-chip companies. Infinity is trying to turn that bottleneck into a product, potentially reducing engineering costs and shortening time to market for hardware vendors.

The bigger picture

The AI-infrastructure market is expanding beyond chips and data centres into the software layers that make hardware usable. As more companies build specialised inference processors, tools that automate compilers, kernels and model optimisation could become essential infrastructure.

The opportunity is large, but execution will be difficult. Infinity must prove that its agent can deliver reliable performance across different chips and models, not just produce generic code that still requires extensive manual tuning.