Infinity, an AI infrastructure startup developing chip-agnostic inference software, has raised $15M at a $100M valuation from Touring Capital, Principal VC, and individual investors including researchers from OpenAI and Anthropic.
Founded by Jeremy Nixon, a former Google Brain researcher, the company is building a CUDA alternative that allows AI models to run efficiently on any chip architecture, from GPUs and phone chips to Systolic Arrays and SRAM.
Addressing The Market Opportunity
Nvidia’s dominance in AI computing rests on two pillars: its chips and its CUDA software stack. CUDA is the low-level infrastructure that allows developers writing in popular languages like Python, using frameworks like PyTorch and TensorFlow, to have their applications run on Nvidia hardware by default. The result is a deep dependency that most AI startups lack the resources or specialized knowledge to break.
Writing custom kernel software, the low-level code that operates chips directly, requires expertise that app-layer companies rarely have in-house. This keeps the majority of AI workloads locked to Nvidia infrastructure regardless of whether alternative chips could offer competitive performance at lower cost. Infinity is building the software layer that would change this equation.
Advertisement
How The Technology Works
Infinity‘s core product is a universal inference library designed to run AI models across any chip architecture without requiring developers to rewrite their applications for each hardware platform. The company’s AI research agent, Ignition, writes the low-level kernel code needed for AI inference on non-Nvidia chips, then tests, debugs, and benchmarks performance automatically, rewriting the code iteratively until performance targets are met.
The system is self-optimizing, continuously learning and improving its output while adapting to different chip architectures including proprietary designs. What could previously take a team of engineers months or years to produce manually, Infinity’s agent can reduce to hours or days. The company does not charge an upfront license fee, instead taking a share of the performance gains and cost savings it delivers, measured in tokens per second.
The approach reflects Nixon’s broader conviction about automated invention, the idea that AI systems can function as a meta technology capable of generating and evaluating new technical solutions in a self-improving feedback loop. Nixon previously developed a machine learning algorithm called Omega that created and automatically evaluated new machine learning algorithms, an experience that led him to apply the same principle to hardware-level software.
Growth And Market Traction
Infinity’s existing customers include D-Matrix, the AI chip company positioning itself as an Nvidia alternative. The company is in active discussions with other chip manufacturers and cloud providers. Infinity currently employs 26 people across engineering, design, and operations, and is backed by investors with direct exposure to the AI research community through its OpenAI and Anthropic angel investors.
Expansion Plans
With $15M in funding secured, Infinity is focused on building out its universal inference library and expanding its customer base among AI chip makers and cloud providers looking for a credible software alternative to CUDA.
Looking Ahead
Jeremy Nixon, founder and CEO of Infinity, described the ambition behind the company’s approach: his obsession with automated invention, and the belief that AI systems functioning as a meta technology can generate the low-level code infrastructure that allows any chip to run frontier AI models effectively. The goal is to make chip-level software development fast enough to keep pace with rapid advances in AI hardware, removing the bottleneck that currently keeps the industry dependent on a single software ecosystem.
About Infinity
Infinity is an AI infrastructure company founded by Jeremy Nixon, former researcher at Google Brain and creator of the AGI House community. The company develops a universal AI inference library and an AI research agent, Ignition, that writes, tests, and self-optimizes low-level kernel software for any chip architecture. Its mission is to build a CUDA-level software stack that is chip-agnostic, enabling AI workloads to run efficiently on hardware beyond Nvidia. Infinity has 26 employees and is headquartered in the United States.
