Anthropic says its AI got Claude running on AMD chips in about three days. The old way took 6 to 12 months.

The claim. Anthropic launched the roadshow for its IPO (opens in a new tab) on Nov. 10. In the roadshow video's "Approach to compute" chapter, co-founder and Chief Compute Officer Tom Brown says Anthropic started using AMD GPUs this year. "In the past, bringing on a new chip platform took dozens of engineers months of work," he says. "This time, we had Mythos try it out. It got Claude running on AMD in just three days, with basically no human intervention. Claude ported or wrote almost 100% of the kernels." (Quotes per Simon Willison's transcript (opens in a new tab).) Kernels are the small, tuned programs that do a model's math on a specific chip.

The fine print. An on-screen chart sets about three days against about 6 to 12 months for the traditional approach. Per The Decoder (opens in a new tab), the footnote limits it to prompt processing for one Anthropic model on AMD MI355X GPUs, at a useful level of performance. That's the stage where a model reads its input. It isn't a whole model, or training, moving to AMD in three days. The 6-to-12-month figure is Anthropic's estimate from its own past chip bring-ups.