NederlandsAnthropic: Claude Mythos got Claude running on AMD chips in three days
According to Anthropic, its AI model Claude Mythos got the Claude models running on AMD chips in about three days, with little human help. Such a switch used to take dozens of engineers months of work.
Anthropic says its most powerful AI model, Claude Mythos, got the company's own Claude models running on the MI355X chips of chipmaker AMD in about three days. The company said this in a video for investors around its stock market listing, which started in the week of 9 November 2026. On Monday 9 November, Anthropic had already released Mythos 5.2 and Fable 5.2, two new models for heavy technical work. Now the company is showing what Mythos was already being used for internally.
A GPU is a chip that does a huge number of calculations at the same time, and AI models almost always run on them. The software that lets such a model do its calculations on the chip is tuned precisely to one type of chip. Anyone who switches to a chip from another brand has to rewrite or adapt a large part of that software. According to an Anthropic executive in the video, bringing a new type of chip into use used to take dozens of engineers months of work.
Mythos wrote almost all the code itself
According to Anthropic, Mythos wrote or rewrote almost all the kernels itself during the test: the small programs that carry out the calculations on the chip. According to Crypto Briefing and DiarioBitcoin, the test took place in July 2026. An Anthropic engineer connected the model to the new hardware and told it to start. After that, Mythos carried on working without constant supervision. AMD chief executive Lisa Su is reported to have confirmed that the extra technical help AMD had ready was not needed.
Anthropic says performance kept improving during the test. The company has not yet published how fast Claude runs on the AMD chips compared with other chips.
Four types of chips
According to the executive, Anthropic now runs Claude on four types of chips: the GPUs of Nvidia and AMD, the Trainium chips of Amazon Web Services and Google's TPUs. This lets the company use computing power wherever it is available. Co-founder Tom Brown said in July that this lets Anthropic choose the best-suited hardware for each task.
On 22 July 2026, AMD and Anthropic announced a large partnership. Anthropic will use AMD chips from the Instinct MI450 series with a combined power consumption of up to 2 gigawatts, in complete AMD server racks called Helios. According to AMD, the first gigawatt is due to go into use in the first half of 2027. AMD also promised to invest up to 5 billion dollars (about 4.5 billion euros) in Anthropic later on. In addition, AMD wants to use Claude to improve its own software for AI chips, ROCm. Anthropic was already using AMD's MI355X chips at the time.
What does this mean for the road to the singularity?
An AI model that writes the software to run on a new type of chip by itself makes AI less dependent on a single chipmaker. If such a switch now takes days instead of months, AI companies can bring new chips into use faster and put more computing power behind their models. It is also an example of AI helping to build the foundations that AI itself runs on. Once Anthropic shares figures and others can check the approach, it will become clear exactly how big this step is.
What is the technological singularity? Read the explanation →
