AI selloff pushes Wall Street into correction
Chip stocks fall hardest as doubts about AI spending grow
Infrastructure
Claude Mythos Preview · IPO roadshow · vs. 6 to 12 months
Written with AI assistance, reviewed by the ACN desk.
“We can onboard new chip platforms in a fraction of the time it once took.”
Anthropic has told IPO investors that its Claude Mythos Preview model got one of its AI models running on AMD GPUs in about 3 days. Tom’s Hardware reports that Anthropic says the work used to take 6 to 12 months. Anthropic says Mythos ported or wrote “almost 100% of the kernels,” the code that runs on a chip. It says this took “basically no human intervention.”
Co-founder and chief compute officer Tom Brown made the comparison in an IPO roadshow video on November 10. He said the earlier effort used a team of “dozens of engineers.” The 3 days covered prompt processing, called prefill, for one model on AMD MI355X GPUs. Tom’s Hardware says Anthropic has not said whether the decode step was included. Anthropic also has not said how the AMD performance compares with Nvidia’s.
Anthropic’s prospectus says: “We can onboard new chip platforms in a fraction of the time it once took.” Anthropic runs its models on Nvidia and AMD GPUs, Google TPUs and AWS Trainium. Its AMD agreement covers up to 2 gigawatts.
Nvidia’s Jensen Huang said in April 2026 that Nvidia uses AI to create its own kernels. He called Anthropic “a unique instance, not a trend.” Tom’s Hardware cites SemiAnalysis, which argues software is no longer as strong a moat as a year ago.
“We can onboard new chip platforms in a fraction of the time it once took.”
Sponsor
Spot an error? Email corrections@aicodingnews.si. Fixes are logged on the corrections page.
Sponsor
Sponsor
Chip stocks fall hardest as doubts about AI spending grow
The Riemann proof passes its formal check; understanding it comes next
Levent Alpöge · proof posted in Lean · not yet verified
The Information: more than $500 million
Sponsor
Sponsor