Standard Kernel Co.

Standard Kernel Co. is a Palo Alto, California-based company developing AI systems that automatically generate highly optimized low-level code (kernels) for hardware accelerators such as GPUs and TPUs. Kernels execute the core computations of AI workloads — matrix multiplications, convolutions, and attention — and, according to the company, their development is a slow cycle of profiling, tuning, and low-level optimization that requires scarce and expensive specialist engineers. The company's stated position is that this bottleneck causes AI performance to lag behind what silicon can deliver, and that automating kernel generation unlocks gains at the model, deployment, cloud infrastructure, and hardware levels, including shortening the time between a new accelerator design and its practical use.

The company's technical approach targets the CUDA plus PTX level rather than higher-level domain-specific languages such as Triton, on the argument that DSLs abstract away low-level decisions, cap achievable performance, and constrain the search space available to an automated system. Working closer to the hardware is intended to give direct access to specialized units such as tensor cores and tensor memory accelerators, which are sparsely documented and typically programmed through inline PTX assembly. An April 2026 blog post describes a hybrid system combining program analysis with LLMs that transforms and optimizes PTX, learning from and combining optimizations observed across DSLs including Triton, TileLang, ThunderKittens, and CUTLASS. Other published work covers GPU benchmarking fidelity (sources of timing variation such as power limits, thermal state, clock behavior, and caching, relevant to RL-based kernel optimization) and the "Standard Kernel Rubric," a framework for evaluating kernel generation systems along five axes: kernel complexity, representation level, hardware specialization, performance target, and automation level.

Standard Kernel raised a $20 million seed round announced on March 11, 2026, led by Jump Capital, and says it will use the proceeds to accelerate development of its autonomous kernel generation platform, expand deployments with AI-native and enterprise partners, and advance adaptive systems software for new models and hardware.