Anthropic moved Claude to AMD chips in three days, almost no humans

Why moving Claude off Nvidia CUDA matters

Nvidia's grip on AI hardware comes partly from CUDA, its software platform for programming GPUs. Code written for CUDA doesn't simply run on other makers' chips, so switching usually means rewriting a lot of low-level software. That is slow and expensive work, which is why a three-day move is the striking part of this claim.

According to Anthropic, that work was done almost entirely without human involvement. The source material doesn't say how the automation worked or which AMD chips were used, so those details are left open.

What it means

For a company that trains and serves a large AI model, being able to shift between chip vendors gives it more room when supply is tight or prices climb. AMD, long the main rival to Nvidia in GPUs, gets a high-profile example of a leading AI lab running its flagship model on its hardware.

The figures here come from Anthropic's own statement. No independent benchmarks or performance comparisons were provided, so how Claude on AMD chips compares with Nvidia in speed or cost is not known.