There’s an interesting bit of news in AI today: In its IPO roadshow materials, Anthropic disclosed that its Claude Mythos Preview model took only about three days to figure out on its own how to get another model running on AMD’s MI355X GPU.
In plain English, that means AI models used to need engineers to spend months painstakingly adapting them to new chips. Now a model can handle it by itself in three days. For the first time, Nvidia’s CUDA moat doesn’t look quite so deep. AMD hasn’t been sitting idle either: its executives have been publicly taking shots at Nvidia and Arm over the past couple of days, saying that four years ago, no one could have foreseen this wave of CPUs “designed specifically for agentic AI.”
My take is that this won’t threaten Nvidia’s business in the short term. You can’t dismantle more than a decade of CUDA ecosystem advantages in three days. But the direction is changing: computing power is shifting from “you can only buy Nvidia” to “go with whoever offers the best value and performance.” AMD and domestic chipmakers are gaining bargaining power. The same goes for AI-themed coins: the narrative is shifting from “Nvidia dominates everything” to “computing power for all.” That’s a positive for sentiment, but don’t rush in. These coins rise on stories, but their falls don’t have to make sense.
As a worker saddled with debt, I only have one thought about this: the savings big tech makes on adaptation costs won’t end up in my pocket. Meanwhile, the value of my little AI-themed position depends entirely on how well other people can tell a story.
How many more years do you think CUDA’s moat can hold?
$TAO $FET
This is not investment advice. Make your own decisions.
In plain English, that means AI models used to need engineers to spend months painstakingly adapting them to new chips. Now a model can handle it by itself in three days. For the first time, Nvidia’s CUDA moat doesn’t look quite so deep. AMD hasn’t been sitting idle either: its executives have been publicly taking shots at Nvidia and Arm over the past couple of days, saying that four years ago, no one could have foreseen this wave of CPUs “designed specifically for agentic AI.”
My take is that this won’t threaten Nvidia’s business in the short term. You can’t dismantle more than a decade of CUDA ecosystem advantages in three days. But the direction is changing: computing power is shifting from “you can only buy Nvidia” to “go with whoever offers the best value and performance.” AMD and domestic chipmakers are gaining bargaining power. The same goes for AI-themed coins: the narrative is shifting from “Nvidia dominates everything” to “computing power for all.” That’s a positive for sentiment, but don’t rush in. These coins rise on stories, but their falls don’t have to make sense.
As a worker saddled with debt, I only have one thought about this: the savings big tech makes on adaptation costs won’t end up in my pocket. Meanwhile, the value of my little AI-themed position depends entirely on how well other people can tell a story.
How many more years do you think CUDA’s moat can hold?
$TAO $FET
This is not investment advice. Make your own decisions.