Anthropic says Mythos cut an AMD chip switch to three days
First merged summary.
Anthropic's IPO roadshow says Mythos ported Claude to AMD GPUs in three days and that Claude already runs on four chip families.
Anthropic told investors during its IPO roadshow that it trains and serves Claude on AWS Trainium, Google TPUs, Nvidia GPUs, and AMD GPUs as one pool. It says frontier model Mythos ported Claude to AMD GPUs in three days with almost no human intervention, versus months of engineering, and wrote nearly all of the kernels. The Decoder says those AMD figures refer to prompt processing on MI355X GPUs. Anthropic also described plans for up to two gigawatts of AMD Instinct MI455X, with the first gigawatt in the first half of 2027, plus Nvidia and AMD investment ties. The figures have not been independently verified, and Simon Willison's transcript was extracted by him rather than released officially.
- Roadshow began November 10; Tom Brown described the multi-chip approach.
- Mythos wrote nearly all AMD kernels; figures cover MI355X prompt processing.
- Anthropic plans up to 2 GW of MI455X; first gigawatt in H1 2027.
- Nvidia committed up to $10 billion and is in a $26 billion placement.
- AMD announced an investment of up to $5 billion in July 2026.
- Claims are unverified; Willison's transcript is unofficial.
Timeline
dated events in the reporting · oldest first- Nov 10, 2026Anthropic began its IPO roadshow · Simon Willison
What each outlet adds
beyond the shared facts- The DecoderAMD kernel claims, MI355X scope, chip plans and investment figures.
- Simon WillisonRoadshow video, Tom Brown remarks, and an unofficial transcript.
Coverage
oldest first · one row per article · UTC- · 6h ago
The DecoderFirst report
Anthropic says AI cut a Claude port to AMD GPUs to three days
Anthropic claims Mythos wrote nearly all AMD kernels and that it already runs Claude on four chip families.
- · 5h ago
Simon Willison
Anthropic IPO roadshow video describes multi-chip Claude training
Co-founder Tom Brown says Claude is trained and served on AWS Trainium, Google TPUs, NVIDIA GPUs, and AMD GPUs.