Startup raises $15m to break Nvidia's AI software lock-in
A new startup has raised $15 million to automate the porting of AI models to rival chips, threatening the software monopoly that secures Nvidia's dominant market position.
Infinity has closed a $15 million seed round at a $100 million valuation to build an AI agent that cracks Nvidia’s software lock-in. Backers include Touring Capital, Principal VC, chip executives, and researchers from OpenAI and Anthropic.
Nvidia holds an estimated 80% of the data-centre AI accelerator market. This dominance is built not just on hardware speed, but on CUDA, a proprietary software layer developed over nearly two decades that makes running models on Nvidia chips seamless. Rival hardware from AMD, Qualcomm and AWS often matches Nvidia on raw power, but lacks this software ecosystem. Porting AI models to these alternative chips requires writing complex, low-level code called kernels, an effort few engineering teams can afford.
Infinity’s product, an AI agent named Ignition, is designed to automate this gruelling process. Human engineers set the direction, but the agent generates, tests, and continuously rewrites the kernel code itself based on incoming performance data.
The company claims early, though unverified, results. Working with chip maker d-Matrix, Infinity says Ignition reached 92% of a new chip’s peak performance just 10 hours after first interfacing with the hardware. Within 10 days, three frontier models were running end-to-end. In a separate test, Infinity reported increasing a model's output from roughly 1,400 to over 20,000 tokens a second in a single day, a rate it says beats the widely used open-source framework vLLM.
For European enterprises and cloud providers, this matters because of the shifting economics of AI. Inference is projected to account for two-thirds of all AI compute spending this year. If Infinity’s technology works as intended, it would allow businesses to run models on cheaper, non-Nvidia silicon, directly eroding the American chipmaker's pricing power over European data centres.
The caveats are substantial. Infinity is a seed-stage company with just 26 employees, a single public chip partner, and self-reported benchmarks. Founder Jeremy Nixon, a former Google Brain researcher, charges a percentage of the cost and speed gains rather than a flat licence fee. “The next era of AI will be defined not just by who makes the best chip, but by who can make any chip run state-of-the-art models at blazing speeds,” Nixon said.
