Nvidia should be sweating if this case is real. Mythos reportedly wrote nearly 100 percent of the kernels in three days to move Claude to AMD.
Everyone knew Nvidia's hardware was expensive, but teams had to buy it anyway. Switching to another chip architecture meant burning months of work across dozens of engineers. Nobody could make that engineering math work.
Now an AI model gets a new chip running workloads in three days flat. That directly smashes the barrier buyers faced when switching platforms. Once teams can calculate compute costs by switching chips at will, Nvidia won't easily protect the premium earned through its software moat.