Early-stage AI infrastructure startup Infinity Inc. raised $15 million in seed funding (post-money $100M valuation) to develop automated software stack generation for AI inference chips. Founded by Jeremy Nixon, a former Google Brain researcher and AGI House co-founder, Infinity's core technology is Ignition, an AI agent that generates, tests, and optimizes low-level compute kernels for inference without human engineers writing them from scratch. The round was led by Touring Capital, with participation from Principal Venture Partners and unnamed executives from OpenAI, Anthropic, and chipmakers.
Infinity's target: breach Nvidia's CUDA monopoly by generating Cuda-level software infrastructure for competitor chips in hours rather than months. The startup is already generating millions in annual recurring revenue through partnerships with d-Matrix and other chipmakers. In a d-Matrix project, Ignition achieved 92% of the Corsair accelerator's theoretical peak performance in 10 hours, distributing Qwen3, Qwen3.5, and Gemma4 fully onto the chip within 10 days. In a separate test on Nvidia H100, Infinity reported 34% inference throughput improvement for Qwen3-8B versus vLLM after one day of automated optimization (not yet independently validated).
Infinity's revenue model ties compensation to performance gains: it typically captures ~20% of the compute savings customers realize from improved throughput, not licensing fees. For architects evaluating alternative AI chip providers (AMD, Cerebras, Intel, Modular/Qualcomm): Infinity's automation de-risks software ecosystem ramp-up, historically the largest barrier to adoption outside Nvidia. The startup's self-improving architecture learns from successes and failures to accelerate subsequent optimization runs. Ignition handles kernel creation, debuggers, profilers, compilers, orchestration across inference hardware—allowing new silicon to achieve usable performance without vendor lock-in to proprietary software stacks.