Infrastructure · 13 Nov 2026
One frontier lab, four chip families
Anthropic described bringing Claude up on AMD hardware. The distinction between first bring-up and a mature production fleet matters. The ability to work across architectures changes who can coordinate the next generation of AI factories.
AWS Trainium, Google TPUs, NVIDIA GPUs and AMD GPUs. Anthropic told investors it could train and serve Claude across all four. Its orchestration layer gives the lab a choice about where a workload runs. That is a strategic asset for a company that describes its most persistent constraint as compute.
The famous three-day AMD example concerned bringing prompt processing for one model up to a useful performance level. Full production qualification is a different clock. Even on that narrower reading, software capable of absorbing some of the work of a port gives an AI buyer more leverage. Losing one influential industry coordinator matters less when the customer has learnt to do some of the coordinating itself.