Lightning network with segwit is decentralized, nodes on datacenters is centralized.
I find this not very convincing. Let us recall what is the centralization pressure in PoW, that has separated the users from the consensus: it are economies of scale in hardware, in electricity. However, these economies of scale are only rising a little bit: if you double your investment in mining gear, you do not double your margin. You augment it with a few percent. This few percent is enough, in the highly competitive market of proof of work, to centralize the PoW stuff so much as to take it out of hands of users.
But let us think now about the LN. Here, the margin is going to be the fee % you are going to obtain on LN transactions, over the price of a settlement on chain. In other words, if you can transact, say, 10 BTC over a channel before you have to settle and "reload", and the LN fee market puts the transaction % at say, 0.1% of the amount, you will win 0.01 BTC of fees, and you will have to pay a settlement on chain, which is maybe 0.008 BTC. You won 0.002 BTC. However, if you can transact, say, 1000 BTC over said channel before having to settle, you will win 1 BTC of fees, while your settlement will still be 0.008 BTC. This time, you won 0.992 BTC net. In order to be able to transact 1000 BTC in a channel instead of 10 BTC, you'd have to commit about 100 times more BTC in the channel of course, if the fluxes are comparable. So with 100 times more investment, you found about 500 times more benefit. Your economies of scale were a factor of 5 ! You invested 100 times more, and your GAIN went up with a factor of 500.
==> you see that the "economies of scale" are not on the percentage level, but are, in our case, 500%.
We see that the LN has a much, much stronger form of economies of scale, namely the concentration of LN hubs into a few big "bank" whale hubs, that can still be very profitable with very low LN fees while smaller fish cannot compete with that. If you cannot commit thousands of BTC into channels, in order to keep them open a very long time before needing to settle, you cannot offer competitive LN fees.
The pressure to centralize by economies of scale in an LN network are much, much stronger than with PoW competition.
But there's still another argument that is dangerous.
The LN derives its security from the ability to settle on chain. However, if the LN contains many more links than can be settled on the chain, then the block chain becomes like "fractional reserve banking". Some can settle when they want, but all cannot settle when they want. If, moreover, the room on chain is scarce, settling becomes an expensive affair, but it also becomes a risky affair if you cannot settle in time.
As such, if there are big LN hubs, chances are they will have bought up room on the block chain in agreement with miner pools, for them to be able to settle if ever they don't like some of their customers. But the other way around will even be more difficult. Settling will not be permissionless any more.
If you consider that mining pools are centralized, but they contain the essential security room for the settlement of the LN, then the LN is even MORE dependent on these pools than simple transaction are, because people's funds are locked in.
==> the LN only makes sense as a decentralized and permissionless system, if it can settle entirely on-chain at any moment. The chain has to be potentially bigger than any LN it supports.