Infinity raises $15M to build a CUDA alternative for non-Nvidia AI chips
Infinity raised $15M at a $100M valuation, led by Touring Capital and Principal VC with OpenAI and Anthropic researchers joining, to build Ignition — an AI agent from ex-Google Brain researcher Jeremy Nixon that writes, tests, and self-optimizes the low-level kernels non-Nvidia chips need to run models. The target is Nvidia's real moat, its CUDA software, not the silicon, and Infinity charges no license fee — it takes a cut of the tokens-per-second gains its code delivers, tying revenue to measured performance. Early customer D-Matrix and talks with other chip and cloud makers mark where the anti-CUDA wave is forming.
Source: techcrunch.com ↗
Infinity doesn't charge an upfront license fee; instead, it takes a cut of performance gains and cost savings, measuring changes in tokens per second.
Why this matters
- → Breaks Nvidia's CUDA lock, enabling chips from other makers to run AI models efficiently
- → Automates kernel writing — weeks of human work compressed to hours or days
- → Revenue tied to performance gains, not licensing — aligns incentives with actual results