Anthropic says its own AI did in three days a job that used to take its engineers months: getting Claude to run on a new kind of chip. The claim, made in the company's IPO roadshow video, goes straight at Nvidia's biggest advantage — the software that keeps most AI labs locked to its hardware.
Anthropic co-founder and Chief Compute Officer Tom Brown said the company's frontier model, Claude Mythos, got Claude running on AMD's GPUs "with basically no human intervention". He said the model ported or wrote almost 100% of the kernels — the small, heavily tuned programs that do a model's maths on a particular chip — needed to make it work.
Three days, almost no human help
Brown said Anthropic only started using AMD GPUs this year. In the past, bringing a new chip platform up to speed kept dozens of engineers busy for months; this time, the company let Mythos try the job instead, and Claude was running on AMD Instinct MI355X GPUs in about three days.
A footnote on the slide in the roadshow video narrows the claim. It says the three days reflects the approximate time for Claude Mythos Preview to bring up prompt processing for one Anthropic model on AMD MI355X GPUs "at a useful level of performance". Prompt processing is the stage where a model reads its input before answering — so the figure does not cover full training or serving.
Brown also said Anthropic now trains and serves Claude across all four major chip families — AWS Trainium, Google TPUs, Nvidia GPUs and AMD GPUs — treating them as one interchangeable pool rather than standardising on a single supplier.
Why it matters — and the fine print
The claim matters because Nvidia's CUDA platform is widely seen as its strongest defence. Most AI software is written for CUDA, and moving to rival hardware means rewriting and re-tuning the kernels underneath — a cost that keeps many labs on Nvidia chips, where hardware margins top 70%.
If an AI model can do that porting work itself in days, the switching cost shrinks — and so, in theory, does Nvidia's grip. That is the argument Anthropic is making to investors ahead of its Nasdaq listing, with pricing expected on November 18 and trading under the ticker ANT the next day.
But the numbers come from Anthropic's own investor presentation and have not been independently verified. Some analysts have also pushed back, arguing it is premature to declare the CUDA moat dead: CUDA remains a vast ecosystem of libraries, tools and years of optimisation that no three-day port can replicate.
A claim that lands as Nvidia shakes
The claim surfaced on a day when Nvidia itself is in turmoil, with chief executive Jensen Huang resigning and the company's market value sliding below $4 trillion. Anthropic's argument is not about Nvidia's leadership — it is about whether the software lock-in underpinning Nvidia's dominance can survive AI that writes its own chip code.
The two companies announced a partnership in July under which Anthropic will deploy up to 2 gigawatts of AMD's next-generation MI450 chips from the first half of 2027, with AMD investing up to $5 billion as milestones are met. Part of that deal includes using Claude to help improve ROCm, AMD's rival to CUDA.
Whether three days becomes the new normal will depend on evidence Anthropic has not yet published. For now, the claim stands as striking marketing from a company about to go public — and a warning shot at the idea that any one chipmaker's software is untouchable.
By Hamza Faisal



