A Call to Humanity: Breaking the CUDA Stranglehold on the Hardware Market

An open letter from an independent researcher who, since 18 September 2026, has directed AI agents to build an independent AI stack on a DGX Spark (NVIDIA Grace Blackwell GB10) in his home. The letter argues that CUDA has become the software layer that decides whose hardware counts in the AI market: a chip that cannot speak CUDA is treated as a chip that does not exist, which holds back rival chipmakers, independent builders, and anyone who wants intelligence running on hardware they own. It answers with receipts: a clean-room replacement for the CUDA user stack on the GB10 (no nvcc, no libcuda, no cuBLAS at run time) that on 22 October 2026 passed a full-chip correctness run of 1,048,576 of 1,048,576 values with the output hash matching the simulator, and Omega, a programming language designed for AI systems to write in, developed in the open. The letter states its limits in full (a qualification queue still open, all measurements on one machine, NVIDIA's open kernel module still in the path, GSP firmware untouched) and closes with a call to action addressed to builders, to the leaders of the open source community, and to political leaders deciding what the next phase of AI looks like: read the receipts, run the stack, break it in public, build on it, and choose owned, open, competitive intelligence on purpose.