Infinity Raises $15M Seed to Make 'Any AI Chip Inference-Ready' — Attacking CUDA Lock-In From the Software Layer

Tech Startups·medium signal

Infinity announced a $15M seed on July 20 for a software layer that lets models deploy on new AI silicon without per-chip porting work. The bet is that the bottleneck to breaking NVIDIA's position is not fabrication but the compiler and runtime surface — the same abstraction play that Infinity's peers in the inference-serving space (vLLM, currently at 86.8K GitHub stars) made for throughput. For anyone modeling long-run inference COGS against the 52% AI gross margins the category is now posting, a viable multi-silicon abstraction layer is the single largest lever on that number.