Infinity Raises $15M to Make Any AI Chip Inference-Ready
Latest | Artificial Intelligence
A San Francisco startup called Infinity has raised $15 million to challenge Nvidia’s dominance in AI, by targeting the software aspect of the chip manufacturer’s advantage.
Nvidia’s Grip on AI
Nvidia’s hold over AI isn’t just due to its fast chips; it’s primarily because of its software, CUDA. This software layer has been the backbone of major AI frameworks like PyTorch and TensorFlow, making it easier for developers to write their applications in Python and run them on Nvidia hardware. As a result, Nvidia dominates approximately 80% of the data-centre AI accelerator market.
Infinity’s Approach
Infinity proposes to loosen this grip by automating the process of porting AI models to new chips. Its AI agent, Ignition, generates, tests, and optimizes the low-level code (kernels) that drives a chip, adapting as performance data becomes available. While the human engineers set the direction, Infinity claims Ignition can achieve up to 92% of a new chip’s peak performance within 10 hours and run frontier models end-to-end in just 10 days.
The Founder and His Vision
Founded by former Google Brain researcher Jeremy Nixon, Infinity is driven by its founder’s passion for "automated invention." This vision involves using AI to create and optimize other AI systems, as demonstrated by Omega, an algorithm that invented and evaluated other machine learning algorithms.
A Crowded Field
Infinity joins a growing number of startups aiming to break Nvidia’s CUDA lock-in. Unlike these competitors, Infinity generates revenue through taking a cut of the speed and cost gains it delivers, rather than charging a licence fee.