Infinity Raises $15M to Build Universal AI Inference Software and Challenge Nvidia's CUDA Dominance

AI infrastructure company Infinity has secured $15 million in funding at a $100 million valuation, with backing from Touring Capital, Principal VC, and researchers affiliated with OpenAI and Anthropic. The startup is developing software designed to help AI chips run AI models more efficiently, aiming to reduce the industry's reliance on a single hardware provider.

The Software Moat Behind Nvidia's Dominance

Nvidia's position as the leading player in AI hardware stems not only from its powerful chips but also from its CUDA software platform. CUDA, which stands for Compute Unified Device Architecture, enables Nvidia's GPUs — originally designed for rendering graphics — to function as general-purpose processors.

The two most widely used AI development frameworks, PyTorch and TensorFlow, have been built on top of CUDA. This means that when developers write applications in popular programming languages like Python using these frameworks, their code runs on Nvidia chips by default.

Most application-level startups lack the resources and specialized knowledge required to write their own kernels — the low-level software that directly operates chips — and port their applications to alternative AI hardware. This creates a significant barrier to entry for any company trying to compete with Nvidia.

Building a Universal Inference Library

Infinity is working to build CUDA-alternative kernel software that functions across a wide range of chip types, including SRAM, GPUs, phone chips, and Systolic Arrays. The company is part of a growing wave of startups attempting to erode Nvidia's market dominance product by product.

The startup's broader vision is to create a universal inference library capable of running on all chips, allowing different hardware platforms to automatically replicate state-of-the-art research results in AI.

From Automated Invention to Hardware

Infinity was founded last year by Jeremy Nixon, a former researcher at Google Brain who also created the hacker network community known as AGI House. Nixon told TechCrunch that he launched the company out of a fascination with the concept of "automated invention" — the idea that AI systems can function as a meta technology.

Nixon had previously developed a machine learning algorithm called Omega, which generated new machine learning algorithms and automatically evaluated them through a feedback loop. That experience led him to consider whether similar automated approaches could be applied to hardware, specifically generating the low-level code needed to run chips more effectively.

The company's AI research agent, called Ignition, is designed to write the low-level code required for AI inference on chips that are not made by Nvidia. The agent tests, debugs, and measures hardware performance with the generated code, and can automatically rewrite the code to improve performance when needed.

According to Nixon, the system is self-optimizing, continuously learning and improving itself. It also adapts to different chip architectures regardless of their proprietary designs. Infinity claims the result is a software stack comparable to CUDA in capability.

Customers, Business Model, and Team

Infinity's customers include D-Matrix, an AI chip maker positioning itself as a Nvidia challenger. Nixon said the company is also in discussions with other major chip and cloud computing firms.

The startup's approach keeps humans in the loop, with people providing high-level direction while the Ignition agent handles the more tedious and repetitive work. In one case study, Infinity found that the agent significantly outperformed human-only workflows, reducing processes that could have taken months or years down to hours or days.

Infinity does not charge an upfront licensing fee. Instead, the company takes a percentage of the performance gains and cost savings it delivers, measured by changes in tokens per second.

The startup currently employs 26 people across design, operations, and engineering roles.

As the AI industry continues to grapple with the high costs and limited supply of Nvidia hardware, companies like Infinity are betting that software solutions can unlock the potential of alternative chips. Whether this approach can truly loosen Nvidia's grip on the market remains to be seen — but the growing interest from investors and chip makers suggests the demand is real. If you found this article insightful, consider sharing it with your network to spark a conversation about the future of AI hardware and software.

Source: TechCrunch