Infinity raises $15M for CUDA-alternative chip software — SkimNews
SkimNews Take
Backing from OpenAI and Anthropic researchers signals that even frontier model labs see chip-software lock-in as a procurement risk worth hedging — inference portability may shift from a startup ambition to a baseline expectation for AI infrastructure buyers.
Get the Tech newsletter
Daily tech — startups, AI labs, chips, the launches that shape the next decade. Free.
- Infinity raised $15 million at a $100 million valuation, backed by Touring Capital, Principal VC, and individual researchers from OpenAI and Anthropic.
- The startup is building CUDA-alternative kernel software that runs across SRAM, GPUs, phone chips, and Systolic Arrays, part of a new wave challenging Nvidia's market dominance.
- Founder Jeremy Nixon, a former Google Brain researcher and creator of the AGI House hacker community, launched Infinity last year, driven by his belief that AI systems can act as a "meta technology" for automated invention.
- Infinity's AI research agent Ignition writes, tests, debugs, and self-optimizes low-level inference code for Nvidia-alternative chips, adapting to different proprietary architectures.
- Chip maker D-Matrix is a customer, and Infinity charges based on performance gains measured in tokens per second rather than upfront licensing fees.
- The 26-person startup claims its agent reduces kernel development timelines from months or years to hours or days, with humans providing high-level direction.
Why it matters: Nvidia's dominance rests as much on CUDA software as on its silicon, and most app-level startups can't write the low-level kernels needed to port to other chips. Infinity automates that kernel-writing grind and ties its fees to measured speedups, giving Nvidia challengers like D-Matrix a practical on-ramp to non-Nvidia hardware.
Ask SkimNews
