A Call to Humanity
Breaking the CUDA stranglehold on the hardware market
My name is Michael Drake Stapleton. I am an aerospace engineer. Since September 18, I have directed AI agents to build an independent AI stack on a DGX Spark sitting in my home. Every claim in this letter carries a public receipt, because that is the rule the work runs on.
The problem
Here is the problem this letter answers. One company's software decides whose hardware counts. A chip that cannot speak CUDA gets treated by the market as a chip that does not exist, no matter how good the silicon is. That grip holds back every rival chipmaker, every independent builder, and every person who wants intelligence running on hardware they own. I intend to break it, and this letter is how I am starting: a working replacement for the CUDA user software layer, built in the open, with receipts anyone on Earth can check.
Where the rented road leads
And look at where the rented road leads. Data centers the size of towns are rising on other people's land, drawing their water and their power, and the companies behind them are already talking about launching more of them into orbit. Intelligence that lives in warehouses belongs to the warehouse owners. Intelligence that runs on your own hardware belongs to you. Competition and free innovation used to be how this field moved. I am putting them back on the table.
What exists today
What exists today is a clean-room replacement for the CUDA user stack on the GB10, with no nvcc, no libcuda, and no cuBLAS at run time. On October 22 it passed a full-chip correctness run: 1,048,576 of 1,048,576 values correct, the hash matching the simulator. The method, the measurements, and the limits are published for anyone to read.
There is also Omega, a programming language designed for AI systems to write in, with its compiler and runtime developed in the open.
Full disclosure
Full disclosure, because a call like this one has to be honest to work: the qualification queue for the GPU stack is still open, everything is measured on one machine, the kernel module is still NVIDIA's open one, and the GSP firmware is untouched. Where this work reaches toward other platforms, that reach is proposal-stage today and is labeled as such.
The call
This call goes out to everyone. It goes in particular to the leaders of the open source community, the maintainers, the foundations, and the people building sovereign, owned tools, because your communities follow where you point. And it goes to the world's political leaders, who are deciding right now what the next phase of AI looks like and who it belongs to. Check the receipts. Test the claims. Tell me where it breaks.
Everything I build is public. Take the work, test it, build on it. Hardware should compete on hardware.
My goals
My goals here are simple. I want intelligence owned by the people who use it, running on hardware they hold, under rules they set. I want a field where any chipmaker, any builder, and any person can compete without asking one company's permission. And I want to work with the people building that world, the labs, the open source projects, the hardware makers, as an independent builder. I'm not trying to work for you. I don't need your money. This is not that kind of letter. I'm trying to partner with you to save humanity.
What I am asking you to do
Read the receipts. The chip run, the paper, the code. Everything is linked above, and all of it is public.
Run it. If you have a DGX Spark, or any GB10 machine, the stack is open. Put it on your hardware and see if my numbers survive your bench.
Break it. File an issue. Publish a failing case. Show me the dependency I missed. A public failure I can fix is worth more to this work than praise.
Build on it. Hardware makers, compiler people, independent builders: the code is there to be used. If your chip sits locked out of this market by a software layer, this is a door.
Lead. If you lead an open source project, a foundation, or a hardware effort locked out by this layer, this letter is addressed to you directly. Bring your project to the table. Test the stack against your workloads. And if it holds, help carry it to your community.
Decide. If you govern, the next phase of AI is being decided on your watch. A field where one company's software picks the winners, and where intelligence lives in warehouses owned by a few firms, is a field your people will rent forever. Owned, open, competitive intelligence is a policy choice as much as a technical one. Choose it on purpose.
Pass it on. Send this letter to one builder who is tired of asking permission.
And write me at drake@aienos.com if you are building in this direction and want to compare notes.
M. Drake Stapleton
AI Architect & Engineer
drake@aienos.com
Cite this letter
@misc{stapleton2026calltohumanity,
author = {M. Drake Stapleton},
title = {A Call to Humanity: Breaking the CUDA
Stranglehold on the Hardware Market},
year = {2026},
month = {October},
doi = {10.5281/zenodo.23614947},
url = {https://aienos.com/call-to-humanity}
}