Inference Startup Infinity Raises $15M from Touring Capital, OpenAI and Anthropic Researchers
Infinity raised $15 million at a $100 million valuation to develop Ignition, an autonomous AI research agent that generates hardware kernels to challenge Nvidia's CUDA moat.
AI infrastructure startup Infinity has raised $15 million in funding at a $100 million valuation to challenge Nvidia's software monopoly. The investment round was led by Touring Capital and Principal VC, alongside prominent AI researchers from OpenAI and Anthropic.
🟩 Breaking Nvidia's CUDA Moat
Nvidia's market dominance in artificial intelligence isn't powered purely by its GPU hardware—it is cemented by CUDA (Compute Unified Device Architecture). CUDA serves as the essential translation layer allowing high-level frameworks like PyTorch and TensorFlow to run seamlessly on Nvidia GPUs.
Because most software startups lack the resources to write low-level chip kernels for alternative silicon, they default to Nvidia:
[ High-Level Framework: PyTorch / Python ]
│
▼
[ Proprietary CUDA Software Stack ]
│
▼
[ Nvidia Hardware GPUs ]
Infinity aims to break this lock-in by providing a universal inference software stack capable of running across diverse hardware architectures, including SRAM, non-Nvidia GPUs, smartphone processors, and Systolic Arrays.
⚡ Automated Kernel Engineering with Ignition
Founded by former Google Brain researcher and AGI House creator Jeremy Nixon, Infinity leverages an autonomous AI research agent named Ignition to write low-level hardware kernels automatically.
Rather than human engineers spending months manually porting AI models to alternative chips:
- Automated Code Generation: Ignition writes low-level kernel code tailored to specific hardware architectures.
- Continuous Self-Optimization: The agent benchmarks performance, identifies latency bottlenecks, and refines code loops autonomously.
- Rapid Deployment: Case studies demonstrate Ignition compressing months of manual hardware optimization down to hours.
🤝 Market Traction and Business Model
Infinity already counts AI chip manufacturer D-Matrix among its active enterprise clients and is in discussions with major cloud providers. Rather than charging rigid upfront licensing fees, Infinity operates on a performance-share model, taking a percentage of client cost savings and tokens-per-second throughput gains.
With a team of 26 employees, Infinity represents a growing cohort of infrastructure startups leveraging AI to dismantle legacy hardware moats and accelerate cross-chip AI inference.
🔗 Reference
- Original Article: Read the full story on TechCrunch
