Breaking

Infinity Raises $15M to Build a Universal Inference Layer Aimed at Loosening Nvidia's CUDA Grip

July 20, 2026 at 11:21 EDT

Funding & M&A
Infra & Chips
AI Agents

Infinity, an AI infrastructure startup building software that automatically generates optimized code for chips beyond Nvidia's, has raised $15 million at a $100 million valuation as it seeks to challenge the dominance of CUDA. The round was led by Touring Capital, with participation from Principal VC and individual researchers at OpenAI and Anthropic, according to a report published July 20, 2026.

July 20, 2026 · Infinity

Taking Aim at Nvidia's Software Moat

AI-infrastructure startup Infinity raised $15M at a $100M valuation, betting that automatically generated kernel code — not new silicon — is the key to freeing AI workloads from CUDA's grip.

$15M

Raised, led by Touring Capital

$100M

Company valuation

People across design, ops & engineering

The d-Matrix case study

Ignition, Infinity's AI agent, compressed kernel-optimization work for AI chipmaker d-Matrix from a human timescale to a machine timescale.

Months → years

Human engineer

→

Hours → days

Ignition agent

How the "universal inference library" works

Ignition writes

Low-level kernel code for the target chip

→

Tests & optimizes

Reproduces SOTA results automatically

→

Runs anywhere

SRAM, alt-GPUs, mobile chips, systolic arrays

Pricing: no upfront license — Infinity takes a share of the performance gains and cost savings, measured in changes to tokens per second.

The bullish case

Attacking the software layer — not the chips — is a pragmatic route to making competing hardware usable at scale, lowering the barrier to alternative accelerators.

The hurdles

CUDA's dominance rests on a deep, mature ecosystem. Auto-generated kernels must match hand-tuned performance — and no release timeline, pricing or public benchmarks exist yet.

Continue reading

The rest of this article is for AI News Blitz readers. Choose an option below to keep reading.