So combining ARM and Ampere operations doesn't hurt any current big customers in the mobile field (apart from perhaps Qualcomm but they're in lawsuit mode there anyhow).
They also make sure to make the ARM alternative available to smaller customers (As an option besides AMD and Intel servers) instead of being gobbled up by f.ex. Oracle (making the big server ARM implementations inhouse to 2 of the big platform operators).
Not to mention the silver lining, combining the ops will give it more credence in the enterprise market that seems to have damn good margins (some have attributed AMD's resurgence and Intel's troubles to Intel slipping in server sales).
And to be fair I think Rene Haas has been fairly upfront about changing their approach.
And all the rest of ARM customers are making mobile/desktop.
So Ampere is going after a completely different segment that the rest of ARM customers aren’t addressing.
Often times these companies might have an idea for how to save power / tweak cores in a weird way on a CPU, but have a chicken-and-egg problem with not knowing exactly how much it'll reduce COGS. Not knowing that and having to negotiate with a vendor (who will often want to charge more for the feature) means that it's difficult to do unless it's an obvious slam-dunk.
By bringing chip development in-house, adding new features skews more towards a political decision that requires less rigorous financial calculations (e.g. "how much power will this save relative to the vendor_cost++ and our own developer cost?" for 3rd-party chips). It basically allows these cloud providers to ship new features more quickly in their CPUs.
I use Hetzner ARM machines myself and they're great, and something something i THINK there's less abuse and noisy neighbors on ARM since it's slightly less skiddie friendly (can't run random x86 binaries)
https://mikrotik.com/product/rds2216
Interesting enough Mikrotik announced a co-development with Ampere last year. https://help.mikrotik.com/docs/spaces/ROS/blog/2024/09/24/27...
There's a distinction that has to be made here.
Different in that ARM only 'designed'. It's more like a framework / library / tool (a very feature complete 1).
The Nvidia case is more similar to having a self-install and then a cloud hosted version of the same software.