Key Takeaways

  • Infinity raises $15 million at a $100 million valuation.
  • The startup aims to create software for AI chips that rivals Nvidia’s CUDA.
  • Infinity’s AI research agent automates low-level code generation.
  • Current customer base includes AI chip maker D-Matrix.

Funding Announcement

Infinity, an AI infrastructure startup, has announced a $15 million funding round, bringing its valuation to $100 million. The investment comes from a group that includes Touring Capital, Principal VC, and researchers from OpenAI and Anthropic.

Challenging Nvidia

The company is focused on developing software that simplifies the operation of AI models on various chip types. Nvidia’s success is attributed not only to its powerful chips but also to its CUDA software, which enables GPUs to function as general-purpose processors. Major AI frameworks like PyTorch and TensorFlow rely on CUDA, allowing developers to use popular programming languages such as Python.

Many startups lack the expertise to create their own low-level software, which is necessary for adapting applications to different AI chips. Infinity aims to fill this gap by developing an alternative to CUDA that can work across various chip architectures, including SRAM, GPUs, and Systolic Arrays. This positions Infinity among a new wave of companies looking to reduce Nvidia’s market share.

Universal Inference Library

Infinity is working on a universal inference library designed to run on all chip types, enabling them to replicate cutting-edge research results automatically. Founded by Jeremy Nixon, a former Google Brain researcher, Infinity is driven by the concept of “automated invention.” Nixon previously developed a machine learning algorithm called Omega, which autonomously created and evaluated new algorithms.

AI Research Agent

The startup’s AI research agent, named Ignition, is tasked with generating the low-level code required for AI inference on chips that are alternatives to Nvidia’s. Ignition can test, debug, and optimize code performance, adapting to various chip designs. This self-optimizing system is claimed to deliver a software stack comparable to CUDA.

Client Engagements

Infinity’s clientele includes D-Matrix, an AI chip manufacturer aiming to compete with Nvidia. The company is also in discussions with other major chip and cloud service providers. Infinity operates on a performance-based model, taking a share of the savings and performance improvements it generates rather than charging upfront licensing fees.

Current Team and Operations

As of now, Infinity employs 26 individuals across design, operations, and engineering roles, reflecting its growth and ambition in the competitive AI landscape.