Bingo. Colocated RAM and the bespoke bridges are a massive part of this too, but buying out what is effectively the entire global capacity for a node is why Apple chips are where they are.

It's a main reason why ipod won so big too, they bought exclusive rights to the only existing small hard-drive tech for music player applications (you could still buy the drives for things like PDAs). Everyone else had to use laptop sized drives.

> buying out what is effectively the entire global capacity for a node

Apple has the money so Apple gets the node which makes Apple more money and everybody else has to wait.

Similar to watching Intel scalp the entire industry, I don't think the US government cares who is in the lead, as long as that company is in their pocket. Apple's move to leapfrog the industry was smart, but also calls into question how actually efficient the chip is. If you compare the Apple A12z (their 7nm predecessor to the M1), to Intel's 7nm offerings, the performance starts weighing a lot closer in Intel's favor. In SIMD benchmarks, Intel can be 10-100x faster than ARM's NEON stopgap.

In other words, I'm really curious what the landscape will look like once Apple, Intel and AMD are all on the same node. I reckon x86 still has some life in it yet, a processor with AMD's IPC enhancements and Intel's process enhancements would cleanly smoke the ARM competition. It just remains to be seen if Intel or AMD will be the first to get there...

I don’t know if their position is quite so secure as Intel’s used to be.

Production can continue to scale up, and there’s massive spending to increase US capability due to the Taiwan situation. Apple obviously still have the money to throw around but I’m sure TSMC want to tread incredibly carefully here with regards to making their products available to more customers. Being an Apple-exclusive fab is lucrative in the short term but encourages your other customers to look for supply elsewhere, which is an existential risk to your dominance.

There are many factors that come together to make the M chips so efficient. The node is one. Another is full integration and optimization of all the systems together. Another is the inherent efficiency of ARM instruction decoding.