No engine can calculate brute force at depth 43 or even close to that. It will use reductions based on probable outcomes which are pretty useless in this particular endgame. Every time an evaluation score is generated other than 99, 0 or -99, the engine admits it may be overlooking critical things - which is effectively through the greatest part of every chess game.
Much worse for the engines than games are artificial compositions. These are set up to demonstrate the power of moves and strategies that are ineffective in common chess games (like underpromotions) and will therefore consistently be misestimated by "game engines" like stockfish. That the engines solve so many of them anyway is due to (a) the tremendous brute force capability that modern engines do have (b) the multi-phased construction of many studies which allows the engines to solve it in "parts"; each part constitutes a tangible success in comparison to alternatives in that part and is within the brute force horizon.This is no different from how they analyze games.
The particular study by Gijs van Breukelen has no clear phases and black seems on top from the first to the prelast move. To the engines this is like crossing an ocean to find "the promised land" on the other side while the stars are clouded and no other navigation tools are availabe. It can send out its probes but few reach far enough to be helpful in plotting a winning course. And those that do have strayed too far to find a way back "home".
Note: "Brute force" is used in a relative sense here since modern engines no longer calculate all the moves in a graph branch. But they will find all the relevant moves and variations within a limited scope by generating enriched random move sequences to locate desirable endpoints (winning positions), and then zoom in on the candidates. Think of it as a fishing net. Nearby the mazes are small and no fish escapes but at a distance they are wider and even a whale might get through. A real world example of the same ideas at work was the "battle of Midway" during WWII where the U.S. aircraft carriers long remained invisible to the japanese reconnaissance net by staying at a considerable distance.
Engines have trouble finding this puzzle because their default settings prune aggressively, especially stockfish.
You let your stockfish run to depth 43. This is great, but it's useless because stockfish already dismissed the winning move. Somehwere down the mainline of the puzzle solution it gave up to early and decided that the line was losing. Those 43 moves are assumming that the first move played is d8=Q.
The way to get engines to get this puzzle and other "anti-engine" puzzles is to reconfigure them (some engines do this easier than others) so that they prune less and are more "open-minded" to seemingly crazy moves. This will let them calculate down the correct lines and see the winning moves.