Segwit transactions are larger than their non-segwit counterparts, due to using more complicated scripting. Yes, the non-witness block space they take up is smaller, but including the witness data they are larger. Meaning they take up more node resources than a simple block size increase would without segwit.
This is the opposite of "scalability". It's making transactions larger, not smaller, taking up more node resources, not less.
Here is a typical segwit transaction, 1 input and 2 outputs https://chain.so/tx/BTCTEST/43c22c3a7f4930684aba86d7fd44ffc42063962e5c32cc662c36b16a7206f018 it takes up 251 bytes.
Here is a typical non-segwit transaction, 1 input and 2 outputs, https://chain.so/tx/BTCTEST/8710532673abe66255f08a47de82d4b05db2e1198ea65e71d8fba1f783e3e4b6 it takes up 226 bytes.
Segwit is being sold as increasing the efficiency of transactions, but in fact the opposite is true. Although it does solve lots of other problems, like quadratic sighashing and malleability. I am still in favor of segwit, because I don't believe node resources are anywhere near being limited.
I never see this fact being discussed in critiques of segwit, and to me it's the most important one.