Infinity's AI chips become inference-ready in days with Agentic tools
Sign in to read and print this article.
Infinity, an early-stage AI infrastructure research company building the software layer that helps AI chips run inference, has released a new case study showcasing its autonomous research and tool-building agent, Ignition. The agent generates, tests, and optimizes the low-level compute kernels, compilers, profilers, debuggers, and SDKs, essentially the foundational software tools, that determine how efficiently a chip actually runs AI models.
The case study centers on a design partnership with d-Matrix, focused on its SRAM-based inference accelerator called Corsair. Infinity says its tools dramatically cut down the time chip companies typically need before their hardware is ready for widespread commercial use. The company frames Ignition as a concrete example of recursive self-improvement, meaning an AI system that builds and independently researches the tools needed for the next generation of AI systems itself.
The underlying challenge Infinity is addressing is a common one in the chip industry: new AI accelerators are frequently held back not by weaknesses in the hardware itself, but by the absence of mature supporting software, exactly the kind of ecosystem NVIDIA has sp...
You've read this far — sign in to keep reading