People have been saying this kind of thing for years, but so far it doesn't really math out.

Having a CPU "in the cloud" is usually more expensive and slower than just using cycles on the CPU which is on your lap. The economics of this hasn't changed much over the past 10 years and I doubt it's going to change any time soon. Ultimately local computers will always have excess capacity because of the normal bursty nature of general purpose computing. It makes more sense to just upscale that local CPU than to rent a secondary CPU which imposes a bunch of network overhead.

There are definitely exceptions for things which require particularly large CPU/ GPU loads or particularly long jobs, but most developers will running local for a long time to come. CPUs like this just make it even more difficult for cloud compute to be make economic sense.

As someone who is using a CPU in the cloud for leisure activities this is spot on. Unless you rent what basically amounts to a desktop you're not going get a GPU and high performance cores from most cloud providers. They will instead give you the bread and butter efficient medium performance cores with a decent amount of RAM and a lot of network performance but inevitable latency. The price tag is pretty hefty. After a few months you could just buy a desktop/laptop system that fits your needs much better.

Larry Ellison proposed a thin client that was basically a dumb computer with a monitor and nic that connected to a powerful server in the mid 1990s.

For a while we had a web browser which was kinda like a dumb client connected to a powerful server. Big tech figured out they could push processing back to the client by pushing JavaScript frameworks and save money. Maybe if arm brings down data center costs by reducing power consumption we will go back to the server.