I could not find anything related.
And this is strange to me, because it seems like it builds packages for Haddock generation.
Hackage is a source code repository only, not a binary cache. One of Haskell's biggest weaknesses is a lack of good binary caching. There are nix enthusiasts, some stack solution, and cabal-cache. It is all rather bespoke an problematic. This is in part because GHC does not have a platform independent IR, but I believe that only scratches the surface.
Processing the source code to pull out the Haddock markup and produce a nice HTML page is a very different process than actually compiling the code to a binary library.
Massively complicating the idea that Hackage could provide a binary cache is that there is simply no such thing as a single binary artefact that is the compiled version of a package. What you get in the compiled binaries depends not just on the source package, but also the platform you're building it on, the exact compiler version, many compiler flags affect the output, and the exact version of every dependency (including transitive dependencies), etc. There are a huge number of possible variations of every package that would need to be built, if we want any package hosted on Hackage to always be downloadable without building.
In particular, if you were pulling a pre-compiled package from Hackage which also depends on package X, and you also use package X in your own source code (or use other packages that also use X), then all of those usages of package X really need to be the exact same version of package X. There will be many different build plans calling for different binary artefacts corresponding to package X. Covering all of them would be a huge ask, and identifying the "right" ones to pre-build to cover many usage scenarios isn't trivial.
So Hackage doesn't even attempt this. It just provides source code and does documentation builds. Inter-package links are a thing, so it's theoretically also true that there are a huge number of possible HTML pages corresponding to the documentation of package X (depending on which exact versions of other packages it links to). But they're not really that critical to match exactly what you're using in your build; the docs are only for human information purposes, and you can always browse different versions of the linked packages manually if you need to.
Stackage providing binary caches would be simpler, since the whole concept of Stackage snapshots is to fix a single version of every package in the snapshot (all mutually compatible, ideally). That doesn't eliminate all of the difficulties, but does remove a huge chunk of them. I don't believe there is any straightforward binary cache of Stackage snapshots either, though.