Anthropic uses AI to onboard AMD GPUs in three days, challenging Nvidia dominance
Anthropic disclosed that it used its own AI model to migrate software to AMD GPUs in three days, a process that traditionally takes six to 12 months. This development reduces switching costs and challenges Nvidia's historical competitive advantage.
During its IPO roadshow, Anthropic disclosed that it recently onboarded AMD GPUs for training using its own AI model, Mythos. Traditionally, migrating models to new hardware such as AMD GPUs required six to 12 months of engineering work due to the complexity of NVIDIA’s CUDA software ecosystem and hardware connectivity advantages. However, Anthropic stated that by using Mythos to write the necessary software, the company completed the onboarding process in approximately three days.
This development significantly reduces the switching costs for AI labs, challenging NVIDIA’s historical moat which relied on the difficulty of porting code away from CUDA. Industry observers argue that AI-driven software porting is eroding NVIDIA’s competitive advantage, making it easier for competitors like AMD to capture market share. Major cloud providers and AI labs are increasingly developing alternative chip strategies to lessen their reliance on NVIDIA, a trend that could reshape the competitive landscape in AI infrastructure.