Standard Kernel

Standard Kernel develops AI infrastructure for automated kernel development and optimization, building AI systems optimized by AI starting with low-level code that executes computations on hardware accelerators like GPUs and TPUs. The company operates at the CUDA+PTX level to retain precise control over architectural details, achieving results like 99-115% of cuBLAS performance for H100 BF16 matmul and 104% of Hopper FlashAttention3 for BF16 attention. Founded in 2025 and headquartered in Mountain View, California, the company has raised approximately $20M in seed funding from investors including General Catalyst and Felicis. Standard Kernel differentiates itself by solving kernel generation at a fundamental level with full hardware control, contributing pioneering research including KernelBench, the first standardized benchmark for AI kernel generation, and demonstrating that LLMs can perform well on memory-bound workloads with proper scaffolding.