Rui Carmo
Another take on the upcoming “switch”, by Ben Thompson. well thought out where it comes to fabbing and core drivers of the transition, but perhaps a tad too optimistic regarding Apple’s role.
For the record, I don’t think the transition to ARM servers really hinges on developers suddenly switching to ARM laptops, if only because I’ve worked with ARM boxes of various sizes for over a decade and it still hasn’t happened yet.
So I find it somewhat amusing to think that Apple would be the catalyst for a “full transition” on the server side, especially given that it is more a matter of overall economics (supply chain, essentially) than anything else.
The truth is that ARM server hardware is exceptionally thin on the ground when compared to conventional servers. A hyperscale provider like Azure might deploy thousands of multi-CPU machines a day (admittedly not constantly), and sources them through massive procurement processes that comply to open, standardized specs like the Open Compute Project.
There is no such kind of standardization (or volume production) for ARM servers, at any level, and even considering that AWS deployed ARM servers two years ago they are far from being hassle free (and, somewhat telling, none of my friends who are on AWS has used them in production yet).
Another case in point is that Scaleway recently announced they would be phasing out their ARM64 servers, which I was testing out last year and which were apparently plagued with maintenance issues.
Again, standardization would have made it easy for them to swap out server hardware.
Switching to ARM is not going to be a “clean cut”, and Intel isn’t going to slowly fade away into oblivion. My money is on a messier scenario with AMD moving up the food chain (which they are already doing, steadily) and two to four more years of ARM server bootstrapping (which is already happening, but very slowly).
Apple may accelerate this motion, but as far as I’m concerned ARM is going to remain an “edge”/mobile platform for a couple of years, and I would be very surprised if it got 10% of the conventional desktop market (let alone the server one) sooner than that.
It’s still a chicken and egg, supply and demand problem, and we’re not going to solve it by having more ducks.