Inference Startup Infinity Raises $15 Million Backed by OpenAI and Anthropic Researchers

AI infrastructure startup Infinity has raised $15 million in a funding round that values the company at $100 million. The investment was led by Touring Capital and Principal VC, with additional backing from researchers at OpenAI and Anthropic. The fresh capital will support the company’s efforts to develop software that allows a broad range of AI chips to run advanced AI models more efficiently, reducing reliance on Nvidia’s dominant CUDA platform.

While Nvidia’s leadership in artificial intelligence is often associated with its powerful GPUs, the company’s success has also been driven by CUDA (Compute Unified Device Architecture). This software layer enables developers to build AI applications using frameworks such as PyTorch and TensorFlow. Because these frameworks are built around CUDA, most AI applications automatically run on Nvidia hardware, making it difficult for competing chipmakers to gain widespread adoption.

Making AI chips easier to use across different architectures

Infinity believes the industry’s dependence on CUDA has created a significant barrier for alternative chip manufacturers. Many AI startups lack the expertise and resources needed to write low-level kernel software that would allow their applications to operate efficiently on non-Nvidia hardware. To address this challenge, Infinity is developing a universal inference library capable of supporting multiple chip architectures, including GPUs, SRAM, mobile processors and systolic arrays.

The startup aims to create a hardware-independent software stack that automatically adapts AI models to different processors without requiring developers to rewrite their applications. By simplifying deployment across various chip platforms, Infinity hopes to reduce one of the biggest competitive advantages enjoyed by Nvidia and encourage broader adoption of alternative AI accelerators.

Founder focuses on automated invention through AI.

Infinity was founded last year by Jeremy Nixon, a former Google Brain researcher and the creator of the AGI House hacker community. Nixon said the company grew from his long-standing interest in“automated invention”—the idea that artificial intelligence can be used not only to solve problems but also to create new technologies on its own.

Before launching Infinity, Nixon developed a machine learning algorithm known as Omega, which was designed to generate and evaluate new machine learning algorithms in an automated feedback loop. That experience convinced him that similar AI-driven techniques could also be applied to hardware optimisation by automatically generating the low-level software required to maximise chip performance.

Ignition automates low-level code generation.

At the centre ofInfinity’ss platform is an AI research agent called Ignition. The system is designed to generate the low-level inference code required for AI chips, automatically testing, debugging and benchmarking the generated software. If performance falls short of expectations, Ignition rewrites the code until better results are achieved, creating a continuous self-improving optimisation process.

According to the company, the software can adapt to a wide range of chip architectures regardless of proprietary hardware designs. Infinity says this approach delivers functionality comparable toNvidia’ss CUDA software stack while reducing the engineering effort typically required to optimise AI workloads for new processors.

Performance-based pricing model

Infinity already counts AI chip developer D-Matrix among its customers and says it is in discussions with several additional semiconductor manufacturers and cloud computing providers. Instead of charging customers an upfront software licensing fee, the company earns revenue by taking a share of the performance improvements and cost savings generated through its optimisation software, measuring gains by increases in tokens processed per second.

Although much of the optimisation work is automated, human engineers remain involved by providing high-level guidance while the AI agent performs the repetitive engineering tasks. In one internal case study, Infinity said work that would normally require months—or even years—of manual optimisation was completed in just hours or days using its automated system.

The company currently employs 26 people across engineering, operations and design as it continues expanding its AI infrastructure platform. With growing investment in alternative AI hardware and increasing demand for software that supports multiple chip architectures, Infinity is positioning itself as one of several startups attempting to reduce the industry’s dependence on Nvidia’s software ecosystem.