Infinity Raises $15 Million for CUDA-Alternative AI Chip Software at $100 Million Valuation
Updated
Updated · TechCrunch · Jul 20
Infinity Raises $15 Million for CUDA-Alternative AI Chip Software at $100 Million Valuation
2 articles · Updated · TechCrunch · Jul 20
Summary
$15 million in new funding values Infinity at $100 million as the year-old startup pushes software meant to help non-Nvidia chips run AI models more easily.
Infinity is building a CUDA alternative: an AI-driven inference library and kernel stack that it says can adapt across GPUs, SRAM, phone chips and systolic arrays.
Founder Jeremy Nixon says its Ignition agent writes, tests and rewrites low-level inference code automatically, cutting work that once took months or years down to hours or days.
The company already counts chipmaker D-Matrix as a customer and is in talks with other chip and cloud groups, charging by taking a share of performance gains and cost savings.
The bet targets Nvidia's biggest moat—CUDA's grip on PyTorch and TensorFlow workflows—as startups try to erode its dominance one software layer at a time.