Every leap in AI capability depends not just on smarter models, but also on the software and hardware infrastructure that underpins them. At the heart of this stack are kernels: the performance-sensitive, low-level routines that execute computations on accelerators like GPUs and TPUs. As workloads scale and diversify, the demand for specialized kernels has surged. Yet the supply of skilled kernel engineers and low-level systems engineers remains severely constrained. For example, maintainer shortages are leaving companies stuck with inefficient code that drives up costs and wastes compute. Production case studies show that kernel-level optimizations can cut CPU and server demand by double-digit percentages, directly lowering compute cost.

Manual kernel optimization has become a critical bottleneck. Each new model architecture or hardware platform requires painstaking engineering, often measured in months of effort. Meanwhile, AI adoption is taking off across industries, creating a widening gap between what's possible with current compute infrastructure and what's needed to fully unlock AI's potential.

At General Catalyst, we back founders building the enabling layers that allow innovation to scale. Just as modular frameworks reshaped cloud deployment and synthetic data transformed fine-tuning, we believe automated kernel generation represents a foundational key to unlock the future of AI infrastructure. 

That's why we're leading the funding round for Standard Kernel, which is building AI-centered infrastructure to automate and continuously improve kernel development.

Automating Kernel Development

Standard Kernel is tackling this problem directly by using AI to generate and optimize kernels. The company's platform leverages large language models and multi-agent systems to automate kernel development, reducing reliance on scarce engineering talent and enabling broader hardware support. By accelerating kernel creation and optimization, Standard Kernel aims to collapse development timelines, expand hardware compatibility, and improve performance across the AI stack.

A Team Built for the Challenge

Our conviction comes down to two factors: the caliber of the team and the clarity of the opportunity. Co-founders Anne Ouyang and Chris Rinard, who met as teaching assistants for Performance Engineering of Software Systems at MIT, combine technical authority with relentless execution. Anne authored KernelBench, an open-source Stanford benchmark for LLM-generated GPU kernels that Nvidia has used in its developer blog evaluations. She previously worked on Nvidia's premier kernel engineering team. Chris brings deep systems expertise and a strong presence as a technical leader. Together, we believe they are positioned to attract top-tier talent from their networks and shape the future of this category.