When people think about mining centralization, they typically think only inside the scope of the single coin. Who are the miners of that coin, and how bad is the centralization there? Those are good questions to ask in Bitcoin, but you only get the luxury of focusing on those questions in Bitcoin because ASICs firmly protect the coin against external hashrate. GPU based coins don’t enjoy these protections whatsoever. When considering the security of a GPU coin, you have to look at everyone who is able to execute a 51% attack, which includes miners and mining pools on other coins, and also includes datacenters and farms that have large volumes of GPUs for other reasons (e.g. machine learning).
It is for this reason that, when talking about Bitcoin security, we usually ignore the raw energy costs of creating a block and instead focus on the active hashrate, and worry about what it would take to get 51% of the hashrate to rally together and attempt an attack against the network. It’s largely unrealistic (even when you consider governments) that there is meaningful hashrate out there which is not actively mining on Bitcoin.
We also know that ASICs are very expensive, and they are exclusively useful for Bitcoin mining. So if you have an ASIC, and you aren’t using it for mining today, you’ve pretty much wasted your money. It’s therefore unlikely that there is significant hashrate in the world that isn’t actively mining on the Bitcoin network — any meaningful amount of such hashrate is a LOT of wasted dollars.