Thursday, 12 November 2026, story 4 of 7
Anthropic says Mythos ported Claude to AMD GPUs in three days
The claim comes from an investor presentation and is unverified. Anthropic says its model wrote or ported nearly all the kernels itself.
What happened
Anthropic says it ported Claude to AMD GPUs in three days, with almost no human intervention. The claim comes from an investor presentation and has not been independently verified, The Decoder notes.
According to Anthropic, its frontier model Mythos wrote or ported nearly 100 percent of the kernels. The Decoder puts the usual effort at six to twelve months. Co-founder and Chief Compute Officer Tom Brown says the company now trains and runs Claude on all four major chip families and treats them as one compute pool. The source does not say whether the AMD deployment serves customers or what it changes in pricing.
What it means
If the figures hold, a large part of the cost of leaving CUDA is engineering time, and that cost is falling. That would give buyers of compute more room to choose chips on price and availability, which could eventually reach inference costs. The source gives no prices, no performance results on AMD and no evidence of savings.
There are reasons for caution. Running after three days is not the same as matching Nvidia's speed or cost. Anthropic says it has invested in multi-chip work for years, so its result may not transfer to teams without such a base. It is also no clean break: the source describes financial ties to both AMD and Nvidia, including up to two gigawatts of AMD GPUs planned from the first half of 2027. Watch for independent measurements and for any sign that this tooling reaches other developers.
What to do: Treat the three-day figure as an unverified company claim, and do not plan around lower inference prices until performance and cost on AMD are shown.
Source
Filed under Chips and compute. Published .
"What it means" is analysis, not reporting. Found an error? Write to hello@vibitely.com.