A young AI infrastructure company called Infinity has secured $15 million in funding at a $100 million valuation, backed by Touring Capital, Principal VC, and researchers from OpenAI and Anthropic. The startup is building software that lets AI chips run models more efficiently, taking direct aim at Nvidia’s decades-long dominance in the market.

Nvidia’s edge isn’t just its powerful chips—it’s CUDA, the software layer that lets developers write AI code in Python and have it run automatically on Nvidia hardware. Most AI frameworks, like PyTorch and TensorFlow, are built on CUDA, locking app developers into Nvidia’s ecosystem. Infinity wants to break that lock by creating a universal inference library that works across any chip—GPUs, SRAM, phone chips, or Systolic Arrays—without requiring developers to write low-level kernel code. The company’s AI research agent, Ignition, writes, tests, and debugs that low-level code automatically, then rewrites it to improve speed.

Founded last year by former Google Brain researcher Jeremy Nixon, Infinity grew out of Nixon’s obsession with “automated invention.” He previously created Omega, a machine learning algorithm that generates and evaluates new algorithms in a feedback loop. That success led him to hardware, where he believes automated systems can write the low-level code needed to make alternative chips perform as well as Nvidia’s. Infinity’s first customer is D-Matrix, a would-be Nvidia challenger, and the startup is in talks with other chip and cloud companies. It doesn’t charge upfront fees; instead, it takes a cut of performance gains measured in tokens per second. With 26 employees, Infinity says its agent can shrink months of human work into hours or days.

Source: TechCrunch