For high-throughput server software x86 is significantly more efficient than Apple Silicon.
In the server space, x86 has the highest performance right now. Yes. That's true. That's also because Apple does not make server parts. Look for Qualcomm to try to win the server performance crown in the next few years with their Oryon cores.

That said, Graviton is at least 50% of all AWS deployments now. So it's winning vs x86.

  ARM in general does not have the top-end performance of x86 if you are doing any kind of performance engineering. I don't think that is controversial.
I think you'll have to define what top-end means and what performance engineering means.

I dont think the point Amazon uses ARM was about performance but purely cost optimisation. At one point, nearly 40% of Intel's server revenue was coming from Amazon. They just figure it out at their scale it would be cheaper to do it themselves.

But I am purely guessing ARM has risen their price per core so it makes less financial sense to do a yearly update on CPU. They are also going into Server CPU business meaning they now have some incentives to keep it all to themselves. Which makes the Nvidia moves really smart as they decided to go for the ISA licences and do it by themselves.