Tech & AI
Anthropic Tells IPO Investors Its Mythos Model Got Claude Running on AMD Chips in Three Days
The claim, made in roadshow material, suggests AI-written code could lower the cost of switching away from Nvidia's software ecosystem, though the scope is narrow.
HedgeWeekly Tech & AI Desk ·
Anthropic told prospective investors in its planned initial public offering that its in-house Mythos model helped get the Claude chatbot running on AMD graphics chips in about three days, with almost no human intervention. Co-founder and chief compute officer Tom Brown made the remarks in the company's roadshow video, according to reports published on November 11 and 12.
Anthropic said Mythos wrote or ported nearly all of the kernels, the small low-level programs that carry out calculations on a chip. It contrasted the result with the six to twelve months and dozens of engineers a new chip platform has typically required. A footnote limits the claim: it covers prompt processing, the step where a model reads an input, for one Anthropic model on AMD's MI355X processors at a useful level of performance. It is not a full migration of training or serving.
Anthropic says it trains and runs Claude on four chip families: Nvidia GPUs, AMD GPUs, Amazon's Trainium and Google's TPUs, treating them as one pool of capacity. In July AMD announced a deal for Anthropic to deploy up to 2 gigawatts of its Instinct GPUs, with the first gigawatt starting in the first half of 2027, plus an equity investment of up to $5 billion. Nvidia, meanwhile, is reported to be among the buyers in a $26 billion private placement tied to the IPO.
Why it matters: Nvidia's CUDA software, which developers have built on for years, is a big reason customers find it costly to switch suppliers, and part of why its gross margin has stayed above 70%. If AI can cut porting work from months to days, rival chipmakers could face a lower barrier to winning workloads. The evidence so far is a single task presented by a company that is courting investors and has reason to stress supplier independence. Watch for independent benchmarks, AMD's deployment timetable and how Nvidia responds as it navigates a leadership change.