A few weeks ago I wrote about how AI-native mirrors cloud-native — lift-and-shift now, real architectural rethink later. The same pattern enterprises went through with the cloud. That post was about workflows and org design. This one pulls on a different, less comfortable thread of the same analogy: Anthropic is starting to look a lot like AWS did a decade ago. Same core move. Much faster clock speed. And I'm not convinced it (or we, startups) survives the difference.
The Core Asset Was Never the Cloud, or the Model
Quick side rant, then I'll get to the point. Anthropic and the other frontier labs are obviously more than infrastructure companies. They do real research, real alignment work, real science. But strip away the mission statements and look at the balance sheet: the core asset is a giant pile of GPUs with a business model of renting time on them. It's just denominated in tokens instead of instance-hours. An EC2 instance is general-purpose compute you point at whatever workload you want. A Claude API call is the same idea, just pre-loaded with one very specific, very valuable workload already running. That's not a knock. Ok, rant over. The interesting comparison isn't the compute. It's what companies do when they sit atop these platforms.
