Preparation has been a thing for a long long time. What hasn't been a thing is the combination of databases and engines. Top players even use supercomputers. The amount of pure grinding has shot up dramatically. Instead of a team of people rattling off lines verbally and debating back and forth, a player can sit at the computer and just hammer out the variations with the computer and see the best engine lines in mere seconds (using a supercomputer).
This is so much more efficient that the only limitation is on how much energy and capacity the player has for rote memorization, rather than how much time it takes for the team of humans to work out the best lines.
More recently I don't think any of the GMs are actually using supercomputers. Powerful computers, yes - but not anything close to supercomputers.
The only two instances I've seen an actual 'supercomputer' mentioned in recent Chess news has been:
* Ian Nepomniachtchi using Zhores (which is questionable given that the institute which runs it [Skolkovo] has Arkady Dvorkovich [chairman of FIDE] as it's chairman). I remember researchers questioning if Zhores was even suitable for the type of calculations which engines care about, so this looks like some PR move more than anything.
* There was a rumour a few years ago that Magnus Carlsen had some secret/private access to a Norwegian supercomputer. No real confirmation of this rumour occurred.
Both AlphaZero and Leela Zero (by their design and very optimal evaluation functions) also require much less evaluation - so they can naturally evaluate further.
Objectively the limits have always been players memory.
Sure, computer analysis has marginally changed how some openings are viewed and sure you can now get an engine evaluation of all the variations of some obscure line in a matter of seconds but novelties remain rare at the top level. Stockfisch is extremely strong because it can play near perfect positional play and can establish micro-advantage through deep calculation not because it fundamentally changed how chess theory is viewed.
AlphaZero might be the sole exception to my point. It did indeed show that modern chess was neglecting some strategical concept but to be honest it was more a rehabilitation of old ideas than a pure novelty.
AlphaZero showed that modern chess was neglecting some strategical concept because engines were. Unlike Go, neural networks play very bold moves but perhaps in a more human than traditional brute force.
But nowadays Stockfish and Leela have both caught up and surpassed AlphaZero.
This is so much more efficient that the only limitation is on how much energy and capacity the player has for rote memorization, rather than how much time it takes for the team of humans to work out the best lines.