Anthropic says its Mythos model ported Claude to AMD GPUs in three days, almost without human help
Co-founder Tom Brown told investors that Anthropic now trains and runs Claude on all four major chip families as one pool of computing power.
In an investor presentation, Anthropic says AI is weakening one of Nvidia's main advantages, the software lock-in created by its CUDA ecosystem. Co-founder and Chief Compute Officer Tom Brown says the company now trains and runs Claude on all four major chip families and treats them as a single, interchangeable pool of computing power. When Anthropic moved to AMD GPUs, its frontier model Mythos wrote or ported nearly 100% of the required kernels, the low-level code that runs on the chips. The Decoder reports that the work took three days instead of the usual six to twelve months, with almost no human involvement. CUDA ties hardware and software together so tightly that many AI companies do not try to switch chip vendors, because porting models takes so much engineering work. If AI systems can do that porting quickly, hardware buyers could gain leverage over chip suppliers and rely less on any single vendor. The claims come from Anthropic's own presentation ahead of its planned IPO and have not been independently verified.