Infinity Secures $15M for Hardware Stack

The landscape of AI hardware is defined by the pervasive influence of Nvidia, whose software ecosystem has effectively standardized how models run on silicon. A new player in the infrastructure space is attempting to disrupt this established hierarchy by automating the creation of low-level software that traditionally requires significant human engineering resources.

Challenging the CUDA Ecosystem

Nvidia’s market position is bolstered by CUDA, the platform that enables general-purpose processing on GPUs. Because major frameworks like PyTorch and TensorFlow rely on this architecture, developers can ensure their applications run seamlessly on Nvidia hardware. Most startups lack the capacity to manually write the kernels required to port software to alternative hardware, which creates a high barrier to entry for new chip designers.

Infinity is building an alternative kernel software stack designed to support various hardware, including SRAM, GPUs, and systolic arrays. By providing a universal library, the startup aims to automate the replication of research results across different chip architectures, potentially lowering the hurdles for hardware competitors trying to gain traction against incumbents.

Automated Code Generation

The company was founded by Jeremy Nixon, a former Google Brain researcher and creator of AGI House. The core of the product is an AI research agent called Ignition, which is tasked with writing, testing, and debugging the low-level code required for inference on non-Nvidia hardware. The system is designed to iteratively rewrite code to optimize performance, adapting itself to proprietary hardware designs without needing manual oversight for every modification.

AI systems can actually be a meta technology.

— Jeremy Nixon, founder of Infinity

Performance and Business Model

The startup is targeting chip manufacturers to help them optimize how their products handle AI workloads. Rather than relying on a traditional licensing fee structure, Infinity utilizes a revenue model based on verified performance gains, specifically tracking metrics like tokens per second to determine their share of cost savings.

• Funding raised: $15 million
• Company valuation: $100 million
• Current employee count: 26

Strategic Implications for Inference

The success of this approach could have significant implications for the AI hardware sector, as it shifts the bottleneck from manual software engineering to automated optimization. If a universal software layer becomes viable, it could lower the switching costs for cloud companies and data centers that currently feel locked into specific vendor ecosystems. For the broader industry, the reliance on automated agents to handle low-level kernel development suggests a future where hardware performance is increasingly dictated by the agility of self-optimizing software rather than just raw transistor counts.