A small AI infrastructure company is trying to solve one of the hardware industry's most stubborn problems: breaking developers' near-total dependence on a single chipmaker's software stack — and investors are betting real money that an AI agent can do what years of human engineering hasn't.

Nvidia's grip on the AI hardware market isn't just about having the fastest chips. Its real advantage comes from a software layer that lets graphics processors originally built for rendering images act as flexible, general-purpose computing engines. The major machine learning frameworks developers rely on today were built directly on top of that software layer, meaning any app written in a popular programming language will, by default, run smoothly on Nvidia hardware and nowhere else.

Most application-level startups simply don't have the specialized engineering talent required to write the low-level code needed to port their software onto alternative chips. That gap is exactly what the new funding round is aiming to close, by building a chip-agnostic alternative capable of running across a wide range of hardware, from custom silicon to phone chips.

The company's founder, a former Google Brain researcher who also built a well-known hacker community focused on advanced AI, says the venture grew out of his fascination with what he calls "automated invention" — the notion that AI systems can function as a meta-technology capable of generating entirely new tools and techniques on their own. He had previously built a machine learning algorithm that could design and test other machine learning algorithms in a continuous feedback loop, and he became convinced the same approach could apply to hardware.

That conviction led to the creation of an AI research agent designed specifically to write, test, and refine the low-level code chips need to run AI models efficiently. The system automatically measures how fast a given piece of hardware performs with its code and rewrites that code when it finds room for improvement, continuously learning and adapting to different chip architectures along the way.

The company already counts an AI chipmaker positioning itself as a challenger to Nvidia among its customers, and it says it is in active discussions with other major chip and cloud providers. Humans remain part of the process, setting high-level direction while the automated agent handles much of the repetitive technical work.

The new round brings in backers including a specialized technology-focused investment firm, an early-stage venture fund, and individual researchers from leading AI labs. The company joins a growing group of startups chipping away, product by product, at the software advantages that have kept one chipmaker firmly in control of the AI hardware market.