Infinity Raises $15M to Build an AI Research Agent That Writes Inference Code for Non-Nvidia Chips

AI infrastructure startup Infinity announced a $15 million Series A at a $100 million valuation on Monday, backed by Touring Capital, Principal VC, and individual researchers from OpenAI and Anthropic. The company is building Ignition, an AI research agent that writes the low-level code needed to run AI models on chips that aren’t made by Nvidia, according to TechCrunch.

The CUDA Problem

Nvidia’s dominance in AI rests on two pillars: high-performance GPU hardware (H100, H200) and CUDA, the software stack that the AI development ecosystem has standardized on. PyTorch and TensorFlow are built on top of CUDA. Developers write applications in Python, use those frameworks, and their code runs on Nvidia chips by default.

Most AI startups lack the resources or expertise to write their own kernels, the low-level software that operates chips, and port their applications to alternative hardware. That software moat keeps customers locked to Nvidia even when competing chips offer competitive performance at lower prices.

Infinity is part of a wave of startups attempting to chip away at that lock-in, product by product, as TechCrunch reported.

How Ignition Works

Ignition is an autonomous agent that writes the low-level kernel code needed for AI inference on non-Nvidia hardware. It tests the code, benchmarks how fast the hardware performs with it, debugs failures, and automatically rewrites the code to improve performance. The system is self-optimizing: it continuously learns and improves across iterations and adapts to different chip architectures regardless of proprietary designs, according to founder Jeremy Nixon.

The result, Infinity claims, is a CUDA-level software stack that works across chip types including SRAM, GPUs, phone chips, and systolic arrays. Humans remain in the loop, providing high-level direction while the agent handles the tedious code generation and performance tuning. In one case study, the startup found the agent reduced what could have been a months- or years-long development process to hours or days, TechCrunch reported.

The Founder’s Thesis

Nixon, a former Google Brain researcher and creator of the hacker community AGI House, told TechCrunch he founded Infinity because he was obsessed with “automated invention,” the belief that AI systems can function as a meta technology. He had previously invented a machine learning algorithm called Omega that created new ML algorithms and automatically evaluated them in a feedback loop. That success led him to apply the same approach to hardware: automated systems generating the low-level code needed to help chips run more effectively.

Infinity does not charge an upfront license fee. Instead, the company takes a cut of performance gains, aligning its revenue with customer outcomes.

Customers and Competition

Customers include D-Matrix, an AI chip maker positioning itself as a Nvidia challenger. Infinity is in talks with other chip and cloud companies, Nixon said.

The timing matters. The AI inference market is scaling rapidly as enterprises deploy agents that require constant model execution. If alternative chips can run the same models at comparable performance through agent-generated software stacks, Nvidia’s CUDA moat weakens. Etched, another inference chip startup, is currently negotiating funding rounds at a $20 billion valuation. The competitive pressure on Nvidia’s software lock-in is building from multiple directions simultaneously.