Infinity Raises $15M to Build CUDA Alternative for AI Chips
Key Points
- Infinity raised $15 million at a $100 million valuation from Touring Capital, Principal VC, and researchers from OpenAI and Anthropic.
- The startup's AI agent Ignition automatically writes low-level code for non-Nvidia chips, reducing development time from months or years to hours or days.
- Infinity uses a performance-based pricing model, taking a cut of cost savings measured in tokens per second rather than charging upfront license fees.
The company's flagship product, an AI research agent called Ignition, automatically generates the low-level code required for AI inference on non-Nvidia chips. The system tests, debugs, and measures hardware performance, then rewrites code autonomously to improve results. It adapts to different chip architectures regardless of proprietary designs, according to founder Jeremy Nixon.
Automated Code Generation at Unprecedented Speed
Read Next
Meta Reports Q2 Earnings Amid AI Investment Concerns
7 days ago
Nvidia Launches Cybersecurity Initiative for Open-Source AI Defense
8 days ago
Ignition's automation capability produces dramatic efficiency gains. In one case study, the startup found that the agent completed work in hours or days that could have taken humans months or years. The system is self-optimizing, continuously learning and improving its output through iterative feedback loops.
Nixon, who previously worked as a researcher at Google Brain and founded the AGI House hacker community, launched Infinity based on his obsession with "automated invention"—the belief that AI systems function as meta-technologies capable of generating new tools for themselves. He previously invented Omega, a machine learning algorithm that creates and automatically evaluates new machine learning algorithms in feedback loops.
Related coverage: Meta Reports Q2 Earnings Amid AI Investment Concerns
Infinity employs 26 staff across design, operations, and engineering. The startup has already secured customers including D-Matrix, an AI chip maker positioned as a potential Nvidia challenger, and is in talks with other major chip and cloud companies.
The company's business model diverges from traditional software licensing. Rather than charging upfront fees, Infinity takes a percentage of performance gains and cost savings, measured in tokens per second. This aligns the startup's revenue directly with customer value creation.
Humans remain involved in the workflow, providing high-level direction while Ignition handles the labor-intensive technical work. This hybrid approach maintains human oversight while leveraging AI's speed and consistency advantages.
Infinity is part of a broader wave of infrastructure startups attempting to chip away at Nvidia's market dominance product by product. The funding round signals investor confidence that alternatives to Cuda-dependent development are technically and commercially viable.
The involvement of individual researchers from OpenAI and Anthropic—rather than the companies themselves as institutional investors—indicates these researchers see potential in Infinity's approach to hardware abstraction. Both companies have pursued GPU procurement strategies that diversify away from pure Nvidia reliance, suggesting alignment with Infinity's mission.
Why this matters: If you're building AI applications, Infinity's tools could eventually let you access cheaper alternative chips without rewriting code—directly lowering your infrastructure costs. For chip makers competing against Nvidia, this software layer is essential to making their hardware commercially viable. For Nvidia, it represents the kind of abstraction layer threat that historically eroded software monopolies, much as web browsers eventually reduced dependency on specific operating systems.
What This Means
Infinity's success depends on achieving parity with CUDA across diverse chip architectures—a multi-year engineering challenge. If successful, it could accelerate adoption of non-Nvidia chips by 2027-2028, fragmenting the GPU market. Failure would reinforce Nvidia's moat and suggest hardware abstraction remains impractical at scale.
Sources: AP, Reuters, ESPN, Bloomberg, BBC and other international news outlets.
Disclaimer: This article is for informational purposes only. Content is based on publicly available news sources.
