Inference startup Infinity raises $15M from Touring Capital, OpenAI and Anthropic researchers

Infinity, an inference startup focused on developing AI infrastructure, has raised $15 million in funding at a $100 million valuation. The round included participation from Touring Capital, Principal VC, and researchers affiliated with companies such as OpenAI and Anthropic. Launched last year by Jeremy Nixon, a former Google Brain researcher and creator of the AGI House community, Infinity aims to challenge Nvidia's dominance in the AI chip market by offering a universal inference library capable of running on diverse hardware platforms. Nixon founded Infinity after his experience with creating a machine learning algorithm named Omega, which generated new algorithms and evaluated them in a feedback loop. Inspired by this success, he explored applications in hardware, believing that automated systems could produce the low-level code necessary for efficient AI chip operation. Infinity's AI research agent, Ignition, is designed to write the low-level code required for AI inference on alternative chips to Nvidia's. It tests, debugs, and optimizes code to enhance performance, adapting to various chip architectures, whether they are proprietary or open-source. The startup's goal is to create a CUDA-like software stack that enables developers to deploy AI models across different types of chips, including SRAM, GPUs, phone chips, and systolic arrays. This approach addresses a key limitation faced by many app-level startups, which often lack the resources or expertise to develop custom kernels for specific hardware. By automating this process, Infinity seeks to streamline the deployment of AI models and reduce dependency on Nvidia's CUDA framework. Infinity's solution involves a self-optimizing system that continuously learns and improves its performance. Users provide high-level directions, while the agent handles the detailed coding tasks, significantly accelerating the development cycle. A case study highlighted that the agent can complete tasks previously requiring weeks or months in just hours or days. Infinity does not charge an upfront licensing fee; instead, it earns revenue through a share of performance improvements and cost savings, measured in terms of tokens processed per second. Currently, Infinity employs 26 individuals across departments such as design, operations, and engineering. Its customer base includes D-Matrix, an AI chip manufacturer positioning itself as a potential competitor to Nvidia. Infinity is also engaged in discussions with other prominent chip and cloud service providers, indicating growing interest in its technology. As Infinity continues to expand, it faces competition from established players like Nvidia and emerging challengers such as D-Matrix. The startup's ability to deliver on its promise of a versatile, self-optimizing inference library will determine its impact on the AI infrastructure landscape. With ongoing negotiations and a clear vision, Infinity is poised to play a pivotal role in shaping the future of AI hardware compatibility and efficiency.