If it’s interesting to you, his public advocacy on his blog for L2 goes back to 2017[1] where he dismisses block size scaling as a centralizing risk. Of course now that we have ofac geth, liquid and cex staking; Parlia Consensus and other consensus methods where the number of validators is capped (usually at a few hundred) but voting is done through delegation are in actual fact more decentralized than ethereum. To be fair though no one knew this is how things would turn out on 2017; back then the idea that just a few hundred actors could control a blockchain was scary. Now it’s a success compared to chains where just 3-5 actors could attack a chain.

[1] https://vitalik.ca/general/2017/12/31/sharding_faq.html