The thesis: a model that writes the chip port itself
Anthropic claims it has cut the porting of its Claude model to AMD GPUs to three days, a task that until now required dozens of engineers for months. The novelty, reported by The Decoder on the basis of an investor presentation, is the method: the operation was reportedly carried out almost without human intervention by the internal Mythos model, which according to the company produced or wrote about 100 percent of the required kernels.
The technical point is that so-called porting does not consist of loading the same software onto different hardware. Each chip family requires dedicated software and inference stacks, and it is this cost that has historically kept companies from switching suppliers. If Anthropic's account is correct, AI becomes a tool for breaking down that barrier, not only for accelerating internal development.
Why it affects Nvidia's advantage
Nvidia's main competitive advantage is the combination of hardware and software, in particular the CUDA ecosystem that ties developers to its cards. Anthropic claims it chose a multi-chip strategy early, when the industry considered it uneconomical, and that it has accumulated enough experience to treat compute from different suppliers, architectures and locations as a single interchangeable pool. According to Tom Brown, co-founder and Chief Compute Officer, the company today trains and runs Claude on all four major chip families.
The consequence indicated by the source is that companies could more easily switch from one supplier to another and choose the most convenient or available option. The material also cites Nvidia margins above 70 percent, described as unusual for hardware, but does not document how any erosion of the software advantage would translate into prices or market share.
Open verification and cross-financial ties
The figures come from Anthropic's investor presentation and, as the source notes, have not been independently verified. They therefore remain company statements, not results replicated by third parties: there is no public measurement of actual porting time, the degree of human intervention, or the quality achieved on AMD compared with Nvidia.
The picture is made more complex by financial relationships. In November 2025, as part of a $45 billion agreement with Microsoft, Nvidia committed to invest up to $10 billion in Anthropic; according to the S-1/A registration statement filed with the SEC on November 10, 2026, Nvidia is also participating in a conditional $26 billion private placement. AMD announced in July 2026 a strategic investment of up to $5 billion, in exchange for which Anthropic plans to install up to two gigawatts of Instinct MI455X GPUs in Helios systems, with the first phase of one gigawatt in the first half of 2027. Anthropic also cites the Anthropic Direct program to manage proprietary capacity and reduce total cost of ownership.
